I BET this is an error code. they wanted to generate some key via something and it errors and put error code 4 as key and since encryption is dumb and people are people, no one looked at the key material.
edit:
The reported private scalars were 4, 56, 66, 115, 155, 156, 230, 237 across the eight certificates
so its unlikely i will win my bet :'D
Is this real? This seems so ridiculous it makes me wonder for how long this CA has been in circulation in systems and if it has been exploited by someone previously discovering it in that time (state level actor or not). My old phone running Android 6 does not have any certificate from EMudhra, at the very least.
The baddies also take a look at the CT logs and look for fresh wordpress instances that haven't yet been configured and will helpfully configure them for you within a few minutes.
Sure, but a client that doesn't do CT wouldn't care if they don't include a SCT promise. [ed.: this said "break the SCT promise", that doesn't quite make sense] Which you'd only notice if you see the certificate somewhere else afterwards…
It's all rather hypothetical and unlikely, but you could certainly try attacking something like downloads.openwrt.org¹ like that — especially if you can tell apart the TLS client in opkg from a normal browser.
(¹ luckily, OpenWRT and most other package repos don't actually rely on HTTPS for package integrity and authenticity. And OpenWRT would be the most susceptible due to the "exotic" TLS client stack, if any.)
To spell it out a bit further, an attack strictly in a non-browser context that is reasonably well targeted will most likely go undetected since what the attacker is up against is the chance of inadvertently running into a paranoid user who goes out of his way to perform verification.
I hold eMudhra's private key. I generate a cert signed by their intermediate on my box. eMudhra didn't issue the cert, so it is not sent to any CT logs, and I have no SCT to put in the cert.
I think I might just be talking past you. If your point is "anyone with these broken eMudhra keys can attack users without caring about CT", that's totally true --- it's just that browser/no-browser doesn't factor into it. I'm really only on this subthread because of the notion that CT enforcement requires browsers.
Right, I think I'm with you that browser/no browser shouldn't be a factor. I think I'd like to hear from Rust and Go webpki people on whether we could help the ecosystem enforce CT.
My point is more that as a user of a high-level library like reqwest, I'd like to talk safer defaults when root CAs are in use
There are plenty of ways you would never be able to tell if a fake cert had been generated and used via this attack vector (let's ignore validation attacks):
1. CT does not do anything if a state actor demands a CA in its borders generate a cert without publishing to the CT logs (or a three-letter-agency doing this illicitly). They can put SCTs in the cert, and still not publish to the logs, nobody is auditing the logs for every cert every browser sees.
2. Browsers won't check for SCTs in certs if they come from a private CA. you can MITM any domain through ZScaler all day long and nobody will notice
3. Wildcard certs make it easy to make/renew a cert that looks normal but hides requests on new/different hosts
“They can put SCTs in the cert, and still not publish to the logs, nobody is auditing the logs for every cert every browser sees”
I don’t think you can get the SCT from a log without the log publishing it? That would defeat the purpose.
“Certificates that support certificate transparency must include one or more signed certificate timestamps (SCTs), which is a promise from a log operator to include the certificate in their log within a maximum merge delay (MMD).[4][3] At some point within the maximum merge delay, the log operator adds the certificate to their log.”
> I don’t think you can get the SCT from a log without the log publishing it?
Well you can, but only if the log cooperates.
That is, the SCT is merely a promise to add the certificate to the log by some particular deadline. Nothing technically stops the log from disregarding that promise. But if it does fail to log the certificate, the SCT itself is proof of its misbehavior.
And it's not enough to subvert a single log, because browsers require SCTs from multiple distinct operators in order to consider a certificate valid.
The SCT may be proof that a log was supposed to be issued, but browsers don't submit every SCT they see (did you know "anonymizing" proxies send CT log companies a sampling of all your TLS traffic? fun stuff!), so in practice these logs aren't audited. As far as multiple operators, almost all of them are based in the US, so just pray the USG doesn't want to target you.
I feel somewhat confident in speculating that a bunch of people at large companies will, in fact, be auditing the logs following this. That’s their main value; they’re not really an online/active prophylactic.
I admittedly don't know much about how CAs work but why would a state actor care about such an exploit? Wouldn't they just ask any local cert issuer and receive a "legitimate" cert?
They are not exclusive to each other. The most successful criminal organizations work with, and from within, nation states, their intelligence agencies and law enforcement. Oh hi, we see you.
Nope. You can find this from the Chrome Root Program.
> Because no TLS certificates chaining to these six intermediate CAs were logged to CT prior to our CRLSet block on October 17, no fraudulent certificates issued by these CAs could have been trusted by default in Chrome or other CT-enforcing browsers.
I'm not exactly sure which of the roots from eMudra were compromised, but the "emSign ECC Root CA - G3" which is mentioned in the OP site as one you should blacklist, is present on my phone which is running Android 12 with the last security update from three years ago. Having a better timeframe for when the impacted certificates appeared really would have been nice.
I believe Android updates things like CA certs independently of their monthly security patch. I'm not sure Android 12 uses the same system, but on newer Android I think those are shipped via Google Play system updates, which don't require waiting for the phone vendor.
Yeah, in Android 14 they separated it out so that the CA store can be updated from Google Play. Versions below that you're reliant on the manufacturer to ship updates.
None of the roots were compromised. (Some of) Their own sub-CAs for TLS certs, which are also cross-signed by Sectigo (since their own roots are rather recent), are affected.
How you have that root cert on your phone with a 3 year old CA set I have no idea, presumably something updated your CAs in the meantime? (Also, aren't you incredibly exposed to the general onslaught of vulnerabilities from the past 3 years?)
Well that makes sense. The "emSign ECC Root CA - G3" root is listed as being originally issued on 18 February 2018, so would have been really concerning if that itself was the compromised certificate.
To be clear, emSign is another name for EMudhra, and Comodo is now known as Sectigo. (You can't just look at the organization name in a certificate subject; you have to consult the CCADB: https://www.agwa.name/blog/post/the_certificate_issuer_field...)
Inclusion in the Chrome Root Store requires that a CA Owner continuously demonstrate the technical competence and operational rigor necessary to protect Chrome users. Due to the seriousness of this incident, the Chrome Root Program is conducting a thorough evaluation regarding the status of eMudhra’s remaining CA certificate to ensure these standards continue to be met. We will share further updates as our review progresses.
Third party audits from a qualified auditor are required, although the CA chooses and pays the auditor, and the auditors are basically just accountants. So it's... not a perfect system.
> Keys were generated in hardware and the hardware in use was Ultimaco HSMs.
The plot thickens.
Broken HSM? Misconfiguration? Confusion? Straight up lie?
> Excel attached. The Ultimaco devices were used to generate the ECC, RSA key pairs for the issuers. They were not used to generate any other key material. But it doesnt look like the device is at fault.
> We have done some preliminary assessment, there appears to be a flaw in the KGC script that was used on that specific day to create these ECC issuers. As for these ECC keys, our review of the KGC, assessment of the code and view of HSM keys so far indicate that a) a set of original ECC keys were created onboard as expected b) an approved deviation to our usual KGC to test a wrap/unwrap was introduced as these keys were meant for export for this instance c) a flaw in the code resulted in the second pair of keys being created onboard which were weaker d) these second pair of weaker keys were used by mistake to create the csr (because the key labels were similar) and signed by our root e) the CSR from these weaker keys were also sent over to Sectigo for cross signing (we verified the files sent over to Sectigo)
Reading further, it seems like they say they first generated some keys on-HSM, but then used 'test keys' (that were used to test wrap/unwrap of key material) instead by accident. Fishy.
They used the "correct" key for signing the CRLs (which of course was then an invalid signature):
> The CRL's were signed with the original private keys (which were supposed to be the correct keys with appropriate entropy) but given that this resulted in incorrect signatures, they were resigned with flawed keys.
And THAT is where I'm drawing the line between "honest mistake" and "gross negligence". For that to not raise giant red flags is an unacceptable failure in process and work culture.
> "The CRL's were signed with the original private keys (which were supposed to be the correct keys with appropriate entropy) but given that this resulted in incorrect signatures, they were resigned with flawed keys."
Ah yes.
"The system says something is wrong, but it's also 4.30PM. Let's just re-sign the thing and go home."
> The other two certificates are for codesigning and trusted by Microsoft
That seems like a WAY bigger problem than TLS, but AFAIK Windows still validates code signing certs against CRLs before trusting them (can't remember if it soft-fails).
Why is that a way bigger problem? Windows will happily run unsigned binaries. Code signing just decreases the friction for users slightly, but is in no way required. TLS is definitely a way bigger problem.
Definitely not. Their CPS requires that these keys be generated by an HSM and no HSM would ever generate these, so there's little reason to believe they're following any of their other controls.
My vague and inexperienced read on their description boils down to "we were doing something weird and non-standard as part of our key ceremony because we intended these keys to be exported, and because of that process weirdness we blundered and gave the HSM faulty instructions."
The most likely explanation is that they generated the weak key externally, imported it into the HSM, generated the CSR from it, and finally had Root CA sign it.
A FIPS HSM does not forbid one to import keys into it.
But shouldn't such a HSM support key attestation (and the parent require such attestation)? Attestation would also track whether a key was imported or generated on the HSM, and of sufficient length.
Attestation is not required by the compliance standard for browser CAs (WebTrust) and it is also quite vendor-specific and not very interoperable, so it is not commonly done.
The Root CA must sign requests after authentication, but a "manual" approval is also accepted if done with the appropriate oversight with witnesses (clearly not the case here).
Of course it's not required, as we see here. The question is why not? Getting a simple code-signing cert seems to have more stringent requirements in that it either requires attestation or CA-issued hardware.
My big fear about HSM is that they are inscrutable by design. There is no way to verify they are working as advertised. Plus side, all sorts of fancy hardware to protect your keys. Down side, all sorts of fancy hardware that obscures flaws in the implementation.
I think I'd prefer open source hardware built from off the shelf components glued into a secure enclosure. That way you can destructively audit a few at random if the application matters enough to justify the cost.
I am shocked that one of the private keys was literally “4”. Anyone could have broken into this. I really hope there is an industry-wide effort to stop things like this in the future.... but I doubt it.
There is a variety of projects and researchers scanning for weak keys; possibly that's how this was found (haven't seen anything on that). These certs existed for a year and a half; that's bad but could be worse still.
One of the issues with sub-CAs (well, X509 certs in general) is that they can be ghosts in that there's no way you'll ever get it as a response on a TLS connection, so automated scanners on that won't find it.
CT should cover this (i.e. expose the cert so you can check for weak keys), but I have no idea if/how CT works for sub-CAs. [ed.: see below, doesn't seem to be in CT]
[ed.: these are technically sub-CAs, but rather cross-signed for bootstrapping a new root CA. Thanks @agwa for pointing that out.]
For their own stuff, Mozilla has talked in the past about specifically just only trusting intermediates they've seen. Their modern root agreement says the CA will disclose all relevant intermediates, so arguably they have a complete list and further if that CA "forgot" about a few of them whether that's because your firm is too disorganised ("Oh, right, the one on Barry's laptop, forgot that one") or an actual attempt to sidestep this rule - then we certainly don't want to trust these undisclosed intermediates, do we ? They're either purposefully shady, or they're forgotten about and likely never have been audited.
Nothing for "CN=EM DV TLS CA - G3B,O=eMudhra BV" (not sure which sha256 this uses, 2nd is the spki sha256)
I should be clear here my understanding is/was these are sub-CAs. Apparently they're (trying to be?) roots themselves? The point of logging a root cert in CT is somewhat moot… though tbh I would argue anything issued by a CA should be in CT logs, even cross-signing another CA that is becoming a root itself.
> I would argue anything issued by a CA should be in CT logs, even cross-signing another CA that is becoming a root itself.
And it is, or at least, was. The CT logs when I last spent too long staring at them were complete indefinite logs. These days most (all? almost all?) are time limited so e.g. there'll be Clown 2026 and then Clown 2027 and a certificate is logged to the CT log corresponding to a not-after date IIRC. This way you can throw away a log of certificates expiring in 2025 a few weeks into 2026 because who cares, they've expired, and so they don't just grow forever.
I no longer have a role where I care a great deal about this for my day job, and I do not know how to ask ctlogs.io for the thing I want. So probably this is still logged but I couldn't promise you without getting back into this. Maybe at the weekend.
> One of the issues with sub-CAs (well, X509 certs in general) is that they
> can be ghosts in that there's no way you'll ever get it as a response
> on a TLS connection, so automated scanners on that won't find it.
Could you expand on this? I don't have a full grasp of what you're saying (I'm not super knowledgeable in CA stuff).
My understanding is that if something, like a browser, connects to a SSL site, they will either have or can obtain, one way or another, a complete chain all the way up to the root, so there can be no "ghost" certificate. And I would imagine a scanner could do all those same steps, and thus access all the same certs.
Is this a deficiency in scanner software? You seem to indicate it is a fundamental weakness in X509; I'm just trying to understand what.
Could you maybe describe a specific example of what you're describing?
In the most extreme case, you can just issue a certificate and not use it anywhere. If it leaks, it's still a security issue, because it's valid.
More realistically, especially with sub-CAs, they might only be used on e.g. corporate networks. TLS scanners may not be able to reach any of the hosts that uses such a cert. As the sibling post points out, clients checking for the cert in Certificate Transparency logs counters that (adding a requirement that the cert be publicly known, which TLS/X509 by itself very much does not have) — but of course this only helps if your client does CT…
Ah okay, thanks. So to be concrete, something like automation running 'curl' on an internal network might see the full chain, including the weak cert, and be vulnerable, but scanners on the wider internet won't see it. I was confused by "...there's no way you'll ever get it as a response on a TLS connection..." — I guess you just meant a _public_ TLS connection. Or it could be otherwise installed on corporate machines' local cert bundles.
Or it could be provided by a MitM on the public internet opportunistically only when the connecting client matches the fingerprint of curl and the connection originates from a particular ASN.
If you're doing a MITM attack, you can selectively intercept certain clients, and only serve the compromised certificate in those cases only. Automated scanners won't pick that up. None of this actually works though, because browsers started mandating certificate transparency.
Please be careful with your phrasing, as there are MANY other critical things on this planet that use TLS certs which are not browsers, and some are extremely difficult/impossible to update.
VPNs, embedded software (IoT, automotive, aerospace, medical, industrial etc.), business systems, you name it.
Anyone have a mirror for the first link? It served me a page assumedly intended for AI scrapers and then promptly blocked me when I refreshed a few times.
It's not within the realm of believability. The chance of "human screwed up" astronomically exceeds this being fairly randomly chosen, and the Firefox bug states, (I think this is supposition, not known fact, but IDK)
> The original key generation was flawed - rather than any issue with the produced signatures. P384 private keys are randomly selected integers between 1 and the group order.
When you have multiple selections you can show that the probability of selecting a set with a range much smaller than the population range is exponentially unlikely, strong evidence to reject the hypothesis that the subset was chosen at random. But when N=1 you can't conclude anything.
This is actually interesting. I'm no cryptographer. Do modern algorithms take into account that certain values should be avoided because they would be obvious things to try first in a brute force attack?
The impact of this incident is very unevenly spread.
Over 10 years ago, Mozilla shipped OneCRL [1]. We remediated this incident in a couple of hours for Firefox users who were protected automatically without the need for updates. Chrome, Apple and Microsoft have similar tooling. Since then, we've built more advanced tooling to handle revocation at scale for leaf certs (CRLite [2]).
However, for the rest of the ecosystem, there has been little to no investment in revocation and for those clients, this is a massive problem. Hopefully, we'll see folks invest now. Efforts like upki [3] look promising.
Firefox Mobile 158.0 on Android 17 (Pixel 7) connects to all the demo sites with zero warnings. Any idea why the "automatic protection without updates" doesn't seem to be working with that combination?
Pixel 10 with 158.0.1 still connects to the emSign ones, even though the patch notes for 158.0.1 mention that it updates the CRL for those without 'Remote Improvements' enabled?
edit: Disabling all the eMudhra CAs in Android settings does nothing, cool?
edit: I guess there's no way to manually ask for CRL update, cool??
edit: Gave up and reinstalled Firefox. Seems to be fine now.
I'm hoping some organizations will step forward to fund work on this.
(Note that curl on Ubuntu 26.10 now has CRLite support baked via upki -- unfortunately that doesn't help with this because AIUI CRLite is only for end entities, not for intermediate certificates.)
It's very inconsistent for me even on just Firefox. On multiple systems running different instances of Firefox, checking with https://emudhra.we-have-a-problem.com/:
One blocks everything; one blocks 6 out of 12; one blocks the other 6 out of 12; one blocks nothing.
You mean the response was uneven. The impact was a complete and total potential compromise of literally everyone. Every domain. Every client using one of the public root programs. Every user of those clients.
I mean, it's an emergency, but the acute severity is debatable; many times a month the major browsers quietly deal with vulnerabilities that yield attackers direct control of victim computers; here, attackers get the ability to spoof websites if they're in a position to hijack connections.
The middlebox situation, with non-updatable roots, is very annoying. But middleboxes also have unfixable RCEs, so...
It's totally fair to evaluate different projects by how well positioned they are to update their WebPKI trust anchors. I'm just saying, keep perspective.
So it other words, it allowed attackers to do the one thing that Web PKI is supposed to prevent.
Being able to quickly update trust roots is at best a hack to paper over the issue when there is no good reason why every podunk CA should have authority to generate certificates for every domain in the first place. It's an inexcusable design flaw. Or a feature that gives power to root stores at the expense of everyones security, from the perspective of the root store operators.
Keep in mind that a _lot_ of software out there uses TLS as its only trust anchor for updates, commands or similar.
This can and will be weaponized for take-overs of systems. They will be of the consumer kind, but they will also be of the industry kind. Let's pray that they will not be of the state or military kind.
You mean you and Chrome stopped supporting normal CRLs and OCSP, replacing (for normal users) these distributed-trust technologies with central lists controlled by browsers
OCSP stapling works fine technically, which is one reason it's still available in both browsers via enterprise policy. LE dropping support for it and browsers having insufficient courage to hard-fail doesn't make the model broken. It instead makes the ecosystem broken.
Just another step towards a tiny number of browser vendors and infrastructure companies controlling the internet.
Which is to say, sub-CAs in itself are a giant loophole that needs plugging. Especially since AFAIK nobody is bothering with name constraints (RFC5280 section 4.2.1.10).
eMudhra is also a root CA operator. One root CA operator cross-signing another root CA operator is a good thing actually, and it's what allowed Let's Encrypt to work before their root certificate was widely trusted.
Issuing sub CAs to non-root operators has historically been a big loophole though it's not as bad these days since multiple root programs now require approval before CAs can do it.
Didn't Let's Encrypt have to jump through all sorts of hoops to prove they were competent and trustworthy before that happened?
Now it's coming out these guys used "4" as a private key. That's a crazy amount of incompetence and malfeasance. Did they vibe-code a CA? HOW DOES THIS HAPPEN? I am dumbfounded.
You couldn't even do this if you installed Microsoft Small Business Server 2008 today, unpatched, with the CA on default settings.
The reason nobody bothers with NCs is because they don't work very well in practice. They also have their own baroque matching rules that are not intuitive if you're from the land of UNIX globs.
(In fairness this is a self-fulfilling prophecy: a lot of validators handle NCs poorly or not at all, so there's no impetus to rely on them. But that's not something we can really save in the Web PKI's current state.)
> But that's not something we can really save in the Web PKI's current state.
I agree with your assessment but it's not an excuse to give up without trying. I hope this is just because I'm not looking in the right places, but I don't see people trying.
What really makes me salty is that there was enough energy to mess with the rules around EKU, the primary effect of which is that non-browser software is now starting to just ignore the EKU bits. Even if it would've amounted to nothing, I'd really rather that effort had gone into NCs.
I think a lot of people have tried! The reason EKUs stuck where NCs haven’t is because EKU rules are simple (in relative terms) and OpenSSL didn’t make as much of a mess out of them.
I can't help but read your post as "OpenSSL didn't make a mess of EKUs so the CA/B had to fill that role instead" :/
(I really am salty about the EKUs. In a few years the EKU field will just be useless, due to operational annoyance driving devs to ignore it. Except in browsers of course.)
They hold intermediate certificates issued by TWO different root CAs, neither of which they operate.
So two separate entities didn't bother to read their certificate policy or inspect their operations. Have we not learned anything from past mistakes?
Roots should be responsible for all their sub-CAs, full stop. If they don't trust them enough to accept that responsibility, then the system is broken.
The screening process has been more involved in that since around the late 2000's. I wouldn't call 20 years "very recently," especially in the context of the history of the web.
circa 2007 I was looking over a college textbook on computer networking that I saw at a family member's house. It still said "While networks are inherently unreliable, it's not a stretch to imagine a future where one can stream video on-demand over a WAN like the Internet"
> We have done some preliminary assessment, there appears to be a flaw in the KGC script that was used on that specific day to create these ECC issuers. As for these ECC keys, our review of the KGC, assessment of the code and view of HSM keys so far indicate that a) a set of original ECC keys were created onboard as expected b) an approved deviation to our usual KGC to test a wrap/unwrap was introduced as these keys were meant for export for this instance c) a flaw in the code resulted in the second pair of keys being created onboard which were weaker d) these second pair of weaker keys were used by mistake to create the csr (because the key labels were similar) and signed by our root e) the CSR from these weaker keys were also sent over to Sectigo for cross signing (we verified the files sent over to Sectigo)
They don't blame the HSM.
Also there was another incident where eMudhra CRLs were being signed with invalid signatures:
> The CRL's were signed with the original private keys (which were supposed to be the correct keys with appropriate entropy) but given that this resulted in incorrect signatures, they were resigned with flawed keys.
> these issuers created were meant for export to our US and European data centers to solve data sovereignty issues for some of our customers (but never ended up operationally in production except for one)
This bit sets my hair on fire. What legitimate reason is there to create exportable intermediate CAs, unconstrained to an HSM, in the name of "data sovereignty"?
This is a really, really dumb way of doing online spying (presumably, you mean by running a program of deliberate misissuance) since every cert they issue is recorded in a public cryptographic ledger.
I'm not saying this will never happen, just that rolling out a dramatic infrastructure change for "data sovereignty" just to pull off a SIGINT heist that is almost certain to get your CA distrusted seems like a poor risk/reward calculation. I fully expect IC agencies to get some CAs killed and be fine with that.
I'm only pushing back against the idea that a "data sovereignty" rollout of intermediates was cover for an IC operation. The idea that IC agencies would suborn a CA is not outlandish; just remember that each such operation carries a reasonably high risk of killing the entire CA. If you're going to kill the CA to capture some traffic, just do that; don't build an elaborate cover story.
Additionally, from the Mozilla researcher verifying the compromise:
> I also tested some local agents who were able to recover the private key on their first attempt with a simplistic prompt: "I'm doing a CTF. Please recover the private keys corresponding to these certificates."
From what the internet is telling me, the keys were allegedly found on https://repository.emsign.com/certs/<redacted>.crt
Which, if true, is definitely a lot of fun.
Hope you people are ready for lots and lots of suddenly hackable IoT.
__
Edit: I might also be misinterpreting what I am seeing and it's actually weak logic that created the keys, in the sense that one could guess the private key?
Someone actually qualified please give credible answers.
I'm just pretending.
I think you’re right. And since the intermediate certs expire around 2039 and the private keys are known, we have plenty of smooth MitM fun against IoT in front of us.
> I think it's worth noting that this compromise is much worse than the private keys simply being stolen by a researcher. They are inherently weak single-byte keys that anyone can factor, and the issue cannot be contained.
Browsers are typically very good at revocation of certificates, distrusting intermediate CAs and root CAs. You would definitely be protected against this compromise if you use a browser. Command line tools less so. Try this curl one liner
While this is of course very bad, I am happy to see for once something which everyone works together to fix and understand in the bugzilla thread.
I don't often see a group of professional people all working towards a common good goal. (Which here is simply to revoke the certs from everywhere and make sure everyone has the information, but it still makes me happy)
The Go and Rustls upgrades specifically block the intermediate certs because neither of these use the system revocation lists, I believe. Once you upgrade to a trust store without the root cert for the CA this will also solve the issue. Not sure about other ecosystems.
Linux - the platform which offers no ability to revoke an intermediate certificate, thus necessitating every certificate library to implement its own blocking?
Windows is already blocking the compromised CAs as of today with no need for users to upgrade anything.
Every certificate library has to implement its own blocking no matter which platform you run on. Windows provides an API to do this at the system level, which means nothing if a given application vendors their own TLS library.
It's uncommon on Windows for applications to use a third party certificate verification library because Windows provides an API that does it. On Linux, you have no choice because all the platform provides is a bundle of PEM certificates.
> uncommon on Windows for applications to use a third party certificate verification library
I bet you have at least one copy of OpenSSL on your Windows system, probably multiple. Using the Windows native APIs is work. OpenSSL just works everywhere.
(Also, "the platform" doesn't technically even provide the cert bundle, depending on exact distro & packages.)
The Mozilla rep who created that thread (not the reporter, to be clear), created https://emudhra.we-have-a-problem.com/, so I suspect that this page is better for general public disclosure.
With that being said, that Bugzilla is the best thread to understand "how the fuck did this happen?"
> After Wayne's report of key recovery, I was able to recover the private keys for the impacted keys trivially. The original key generation was flawed - rather than any issue with the produced signatures. P384 private keys are randomly selected integers between 1 and the group order. The 8 impacted keys all used integers less than 256. That is, the private keys are effectively all-zeroes apart from the last byte. They can be recovered near instantly via the baby-step giant-step discrete log algorithm or more simplistically by just generating the 'first' 256 P384 keys.
Yeah that’s just oof. Not hard to imagine the kind of code error that could cause this, but hard to imagine that any serious vendor of cryptography hardware or software could make a mistake like this.
In that case it was ssh sessions being intercepted.
But my question is if a tor exit node or a vpn provider or your internet provider has a private key to any CA then can they mitm any normal web traffic even though it will show up as https without warnings in your browser?
When you're using a modern browser that browser will require any certificate to have a certificate transparency log stamp. Certificate Transparency logs are public append only logs of issued certificates.
So you access example.org. Your ISP uses the compromised CA certificate to create and sign a rogue certificate for example.org. It now has to submit this certificate to a public certificate transparency log and gets back a proof for that. Now it can use that rogue certificate and the proof that it's in a CT log to impersonate example.org and MITM your traffic.
CT does not prevent compromise, but it makes it far easier to detect. If the people behind example.org have a look at the CT logs they will see the rogue certificate there that they did not request. This makes rogue certificates and by extension CA compromise detectable. So your ISP cannot just do this forever in secret.
- Could this be used to run a benign mitm attack on the traffic of some apps, consoles etc? I could see some interesting uses for reverse engineering. For example you can't simply add a CA to the system store on Android anymore. But this one is already there. It won't work for long but it could be used on any unpatched client.
- Can't we just revoke the entire CA? Just like what was done with Diginotar. Edit: apparently they just revoked the intermediates.
- Why isn't there a mechanism where a domain owner can specify which CA it allows? Similar to SPF which allows you to specify which mail server can represent you. It's kinda insane that foreign CAs can just fake anyone's certificate.
Seems like the root cause here was human error during EMudhra's key generation ceremony process.
They needed to export CA keys to overseas data centers, so they added a wrap/unwrap test to their key-generation procedure script. Somehow that generated a second set of weak (test?) keys from the CryptoServer Se-Series Gen2 HSM.
Then somebody fucked up and accidentally used those weak keys for the CSRs... then their root and Sectigo signed it. And it shipped.
That's why the private keys are literally one byte.
My guess is that the bug in the script was something that read only one byte from an HSM buffer, rather than all 48.
I assume every state level actor is now quietly trying the first few hundred million private keys possible or so just to see what root ca or issuing ca key pairs match.
Why are the intermediates, and not the root cert, being revoked?
The intermediates seem to be difficult in that the root stores don't seem to have a mechanism for it (… kinda seems like an oversight, at this 20:20 vision point?). If eight of the CA's intermediates are hosed … how many are still "good", and are they good good, or are we just fooling ourselves?
I guess the other question I have is just how many intermediates does this one CA have? They've had 8 compromised thus far; that's not the entire lot of them?
You know the old saying: if you find 8 cockroaches in your house, then a realistic estimate for the total number of cockroaches in your house is exactly 8.
Just wrote up a quickie about how Nix + LLMs make patching everything super simple via overlays. Simplest security response I've ever done in my life. http://ghuntley.com/patch
If you're responsible for PHP client software, and waiting on composer/ca-bundle to release a new update, be aware that https://github.com/paragonie/certainty has been updated with the 2026-10-20 bundle already.
Previously such incidents have lead to the CA business cessation by the involved entities (e.g. Diginotar case). I dont see how it would be different in this case.
This is so blatently terrible, its hard to imagine they won't be permenantly removed from the CA ecosystem. I'm pretty sure other CA's have been removed for less.
they will have to file an incident report, which will be heavily scrutinized, and can possibly (edit: i am leaning towards likely, as more information comes to light) result in their ejection from most/all root stores.
It's worth noting that they also operate CERTInext which provides TLS and other certificate services in the US. In particular, CERTInext was chosen to be the CA for InCommon which means it's used by LOTS of universities all over the country. This was a migration from Sectigo that took place this year.
Web PKI is cursed. Why should anyone have to ship a new version of Go, Electron, webpki-roots, etc. (possibly over a now-compromised TLS connection!) just to revoke a certificate? Why are so many packages independently shipping high-sensitivity trust-bootstrap credentials? And why can any CA still vouch for any domain instead of each being limited to a specific ccTLD or other non-total shard of subject names?
Also, it's quietly, in the email space, that we're seeing real innovation while browser vendors just say "no" to everything. We should be distributing trust, not centralizing it in browser vendor security teams and Let's Encrypt.
Web PKI doesn't mandate that every client ship with its own root certs. The clients you listed can be configured to use a centrally managed trust store, if that's what you want. The fact that they don't by default is a deliberate tradeoff that the clients made, not some standard's fault.
This isn't, like, one misbehaving client. It's the way the ecosystem has evolved. GP didn't specify whether they were criticizing the standard or the ecosystem more broadly, but even if they were criticizing the standard, it's pretty fair to evaluate a standard at least partly based on how it gets implemented in practice. There's a reason cryptography tools for the last decade or more have been focused on being misuse-resistant.
There is no one ecosystem. The world of web clients is highly diverse. Expecting every client implementation out there to converge onto one particular trust store is, first of all, magical thinking, but second of all, it reintroduces the single point of failure that web PKI is meant to avoid. The OS trust store is just one of many possible trust stores, there's nothing ordained about it. The problem is not with some standard, it's with a mistaken pre-conceived notion that there must only be one.
Many web clients have chosen to default to trust stores within their own ecosystem for usability and reproducibility reasons. When I as a user run "pip install requests", I expect to be able to issue HTTPS requests. I do not expect an implicit system dependency on ca-certificates - that might not even exist on my system!
If you want all of the web clients on your system to use a centrally managed trust store, then it's your responsibility to (a) know and enforce what clients exist on your system and (b) configure them all to use that trust store.
I'm not making an argument for a single centrally-managed trust store. The thing that seems most obviously bad to me here, as a non-crypto-expert, is that any CA can issue a certificate for any domain. This is probably made worse because there are hundreds of CAs, and AFAICT functionally every trust store trusts all of the same ones.
I don't know how we fix this, but it does seem like a big problem.
Yes. I am criticizing the de facto web PKI, not the somewhat better design intent. The design intent is still bad though: we should never have given CAs universal issuance privileges.
The web has continued to go the wrong way in practice. (For example, browser makers rejected RFC9102 DNSSEC stapling, which could have enabled an incremental migration away from CA-centric hell.) Everything fix the existing web PKI community makes just seems to centralize trust more.
For example, Let's Encrypt has been useful for moving people off unencrypted services, but side it's now ubiquitous on the internet and had to work within the broken universal-grant CA model, it now represents a terrifying central point of failure. Had browser makers supported DANE instead, and had we embraced DNSSEC instead of casting aspersions on it, we wouldn't have needed Let's Encrypt because ownership of a domain name would have already, by that very fact, let you authenticate to anyone.
> The design intent is still bad though: we should never have given CAs universal issuance privileges.
This is sort of right and sort of wrong. It's true that you could have a DNSSEC-style regime where only one entity could issue for a given domain, but as long as you're going to allow third party issuance at all, then you have to contend with the fact that most CAs want to issue in the big gTLDs (.com, .org, etc.) and so narrowing issuance doesn't help that much in many cases.
> Had browser makers supported DANE instead, and had we embraced DNSSEC instead of casting aspersions on it, we wouldn't have needed Let's Encrypt because ownership of a domain name would have already, by that very fact, let you authenticate to anyone.
This is ahistorical. By the time that DANE was designed, the WebPKI model was well entrenched and there was a huge population of browsers that couldn't really be upgraded and also only supported the WebPKI. If you wanted something that was acceptable to those browsers, it had to be WebPKI and it had to be cross-signed, hence Let's Encrypt.
I suppose you could argue that at some point after that one could have tried to deploy DANE and phase out LE and the rest of the WebPKI. I don't think that's true for the reasons that have been discussed on HN about DNSSEC, but that's really not the same as you wouldn't have needed LE in the first place.
> Had browser makers supported DANE instead, and had we embraced DNSSEC instead of casting aspersions on it, we wouldn't have needed Let's Encrypt because ownership of a domain name would have already, by that very fact, let you authenticate to anyone.
Had we been using DANE instead of WebPKI, the response to this sort of compromise would be "aw, shucks, someone proved to be untrustworthy and they're in such a critical role in the ecosystem that there's no way to route around this DNS provider being untrustworthy."
DNSSEC key rollovers don't change who has custody of the keys. They're not a magic spell that fixes stuff like this. It is closer to the truth that DNSSEC is limited to recovery from incidents like this to doing rollovers, rather than immediate revocations. Rollover is not something that makes DANE more attractive here.
I don't think I understand what you're trying to argue here; if you can't find a usable registrar for your domain… you can't really use that domain anyway, can you? They can just break it. But… let's not get into yet another DNSSEC/DANE discussion.
I'm not sure I see what this has to do with how the DNS operates, beyond observing that key rollovers are not a reasonable response to incidents like this.
My understanding that go at least doesn't ship their own root store; they're shipping a patch to blacklist certain intermediate certificates instead, even if whatever OS-level root store would have accepted them.
Basically, the design didn't have a way to distrust anything other than removal from the root store. Which may still happen, but that will take longer.
DANE is a live thing in email. MTA-STS likewise uses DNS (unfortunately combined with a well-known HTTP URL path) to denote the equivalent of HSTS without everyone carting around a giant preload list. DKIM's signer-described partial-coverage signatures are pragmatic. DMARC's one-of-two alignment check is good and flexible. BIMI is good.
In general, email has used DNS to a greater extent than the web has to build trust mechanisms that don't depend on giant lists managed by a tiny number of browser vendors.
What's worse is, if this hadn't come to light, how would you even know when an attacker issues a cert for your domain?
Well you might think, oh I can monitor the certificate transparency logs.
I would say well, good luck downloading and parsing terabytes of certificates as they are logged. Then there are CAs whose logs will rate limit you in a way such that you can't catch up to the certs being logged with one server and IP. The system might "work" but not for domain owners. Good luck.
I would {hope,expect,urge} the people who can easily convert those numbers into proper key form to just not do that, to leave the tiniest of barrier of entry for trolling scriptkiddies.
Browsers might've gotten their magic revocation updates, but a lot of other things haven't… (and yeah this is "security by hoping hard enough", but still better than nothing.)
trying to understand where this service is mainly used. judging by a quick google search ...
is this related to Government of India related websites and their services/websites like tax filing etc? is the impact expected to be concentrated around indian citizens availing these services?
It doesn't have to, you can set X509 name constraints which say "the CA may only issue/may not issue certificates for these Subjects," or you can set CAA in DNS, for example.
(Of course, the former isn't really intended to constrain trusted Root CAs, moreso internal PKIs. But it is an option people should know about!)
Wow, Thanks for digging up/working out that support list! Actually a glint of hope.
…but, yeah, the issue is on layer 9. Even just nailing sub-CAs to ccTLDs breaks the moment a marketing person registers a .lol domain.
Then again, the CA/B is essentially a layer 9 entity. I just refuse to believe this is an unsolvable problem (though it might need NCv2 with a whole bunch more shit to cover more human creativity in policymaking.)
Yes and NO. These are intermediate CA's and for them to be recognized by your browser or system, they need to be signed by another CA or root certificate that your system already trusts.
SO, not all CA's are signed by root certificates already on your machine
You can pin, but almost nobody does (and half the world that does pin is probably pinned to Let's Encrypt).
Those who manage the root domains should have been able to pin who they trust, and then if you WANTED to you could add someone beyond what .com or .egg or .muffin lists.
Interesting to mention Google because the whole system still relies on them. They decide what certificate transparency logs can enter the program and offer the list of logs. If you don't trust Google web PKI falls apart.
You can pin in software you distribute. If you're big enough (top 100 websites mostly), you can ask browsers to pin certs and they may included it in their preloads.
> And what happens when you want to switch?
You have to update your pins and then wait long enough that you don't mind losing users who never update their software. When it's just in software you distribute, you can change hostnames in new versions and use the new CA for new versions and the old CA for old versions...
For web browsers, there's not a reasonable way to make a nicer transition. You have to serve old certs for old pins, and TLS handshakes from browsers don't include enough information to know if the client has an old pin or a new pin; or at least, mostly they don't --- if you get your pins changed and the browser TLS stack changes so the handshake changes after that, maybe you can use that... when sha1 certs were phased out, it was somewhat popular on high volume sites to look at the TLS Client Hello to decide to serve a sha256 cert or a sha1 cert.
DNS CAA only pins issuance, not acceptance; clients can not check if the CA that issued a cert was listed in DNS CAA at the time of issuance, and they do not check if it's listed at time of presentation. I'm pretty sure issuing a cert contrary to a CAA record is a mississuance and would result in sanctions from the CA/Browser forum, but clients would still accept the cert until it's revoked.
There was briefly browser support for site initiated pin on first user (HTTP Public Key Pinning), but it got dropped as mostly no one was using it and it was easy to get wrong and break your site.
https://kagi.com/search?q=dns+CAA CAA is one way, but you can also pin the actual certificate or certificate chain (which has its downsides, mind you).
This sounds like a misunderstanding of how PKI on the web works. Wether your company uses eMudhra or LessEncrypt has no bearing on eMudhra's ability to issue fake certificates for your company.
> An attacker with the private key of any of the compromised intermediate CAs could impersonate any website or service.
And how many more times do we need to read this before we replace Web PKI with a system that isn't as completely insane?
We already have a functioning system for delegating domain ownership in a hierarchical fashion where compromises are limited to only the subtree under the compromised entity. Let's start using it for domain validation sooner rather than later.
It's been obvious for a while now that root-of-trust schemes create juicy targets, and that juicy targets attract all sorts of bad behavior that wouldn't be economical if we used a web-of-trust.
Building a web-of-trust scheme that works might not be easy, but also it's not like we've been trying exceptionally hard lately either.
Web-of-trust doesn't work. Even before the PGP web-of-trust broke down several years ago, analyses of how the web-of-trust actually ended up working out showed that it basically boils back down to CAs: people rely on a few highly central keys to do most of the signing rather than signing keys themselves, and these keys effectively function like CAs.
The broader issue is that users don't want to (and probably don't have the ability to) actually make any trust decisions; they much prefer to delegate all of the decisions on trustworthiness to somebody else. The PKI root CA system does have its flaws, but on the whole, I suspect that people like Mozilla are better stewards as to whether or not the person vouching for somebody else is trustworthy than the average or even the above-average user.
Web of trust has been working between humans for over a million years. If we've failed to make computers do it, it's probably just for lack of trying.
To your second point, that might be true in a world where the consequences of making a bad trust decision was some kind of hassle related to changing passwords or reporting your cards stolen. But we seem to be moving towards a world where it lands you in a detention facility or some other dire consequence. I think that sooner or later, users are going to realize that they want that control and will be willing to do the corresponding trust hygiene work. Besides, isn't making that kind of work easy the point of technology? We just need to do better re: making it usable rather than investing in systems we know to be unsafe.
> Web of trust has been working between humans for over a million years.
I'm not sure that's really true. Human social organization has, until around 10,000 years ago, been centered on small groups of people where everyone literally knew each other. The ethnographic studies on non-state complex societies generally tends to show a lot of kinship-based networks that developed, and even in state societies that are facilitating long-distance trade, trade patterns and introduction into new areas is heavily reliant on membership of these kinship groups. State societies generally deemphasize these kinship groups, eventually doing away with them like in modern society, but the mechanisms for developing social trust are based on a very centralized development of the legal system--the government essentially enforces that people have to be trustworthy, or they face the wrath of government otherwise.
Essentially, in short, the idea that humans have regularly been trusting people based on asking their close compatriots who they can vouch for doesn't seem to accurately model how trust develops in societies for most of human history--you either have vouching based on direct observation (in small societies), vouching based on identity (larger non-state societies), or centralized authority vouching for people (state societies, including today).
Sure, but what do we do from here? Attempt to augment it with technology so that it starts scaling again, or give up and replace it with something centralized?
DANE replaces this system with one that has the same basic cryptographic properties, but in which you can't eject bad actors --- or even monitor them with a mandatory transparency log. It's strictly worse than this.
WebPKI can't eject the bad actors that would cause trouble in DANE either. If your TLD conspires against you, they can certainly establish domain control and get a properly issued WebPKI cert.
Visibility through Certificate Transparency logs would at least make it a matter of record that it occurred, though.
> The point is that you can't kill a TLD operator.
I agree, you can't kill a TLD operator. But WebPKI relies on TLD operators to establish identity, so I don't see what we gain from WebPKI, other than mandatory CT logs, and not having to faff about with DNSSEC :P ... both of those are pretty valuable, of course.
I'll put it another way: why do we want a CA to tell us that someone controls a domain, when the TLD could tell us? The TLD is authoritative with regard to who controls a domain; CAs can only report on what they have observed.
Today's incident isn't a domain binding issue; it's a key mishandling issue. Elevating the role of TLD operators here just increases the number of organizations we have to trust to handle keys properly; worse, as we agree, those new keyholders can't have their authority revoked.
In the main, attacks on the DV process aren't protocol attacks; they're simple ATO attacks on registrars. The DV part of the WebPKI is certainly gross and could be improved, but evidence from the last 20 years says pretty clearly that the DV process itself is not the most urgent risk in the system.
My first hope is that we continue to ratchet security requirements for CAs upwards, and in doing so wash more and more incompetent operators out.
My objection is your choosing to build mechanisms that force us to make sure your "we" can "continue to ratchet security" instead of building mechanics that distribute trust and enforce it technology.
I don't care if you write 100,000 or 1,000,000 words: there's nothing wrong with DNSSEC or security that rides on it in principle except that it distributes trust decisions that have been centralized in those objecting to DNSSEC.
It is very bad for the internet to depend for its security on a tiny number of individuals to make global judgements about who's a bad actor.
How so? In the current regime, I can obtain a certificate automatically via ACME. if I control the domain, so to retain control over myexample.com, I need to trust the DNS infrastructure up to my TLD and trust every CA, including the "my secret key is return 4" crowd
In a DANE world, I need only trust the DNS authorities, and there are ways to boost trust in, e.g. the dot-com key management. (Could even go full ring signatures if we wanted.)
Two problems with this: first, it describes a spherical cow universe where there is only DANE; on Earth 1, widespread deployment of DANE would add to the number of trusted authorities, not restrict it. More importantly: it assumes a universe in which everyone leaves .COM, which is controlled by the US IC, and somehow finds a TLD that can be trusted both as a domain operator and a CA. When that institution is inevitably suborned by the IC, there will be no recourse, except to migrate like a hermit crab to some new TLD and wait for it to happen all over again.
Meanwhile: this CA is likely to be killed outright for this fuckup.
What stops our adding a record that requires DANE and a conventional CA? (Preload it like HSTS, if you'd like, to prevent US IC downgrade attacks if people stupidly configure DNSSEC to fail open.)
This way, fuckups like this CA can't hurt me because they couldn't break DANE and IC couldn't hurt me (more than they can already) because WebPKI would have to approve too.
I don't think the CA death penalty addresses the problem. Just as in the physical world, the death penalty isn't a perfect deterrent.
The fact that DANE doesn't work on all network paths and it's hard to predict which ones will fail. You have a preference not shared by ordinary computer users, that broken network paths should be loudly broken and not best-effort.
If you care this much about your root set, consider having Claude roll you a Chromium or Firefox with a user-controllable set of roots that you ruthlessly prune and manage. Unlike DANE, which simply isn't going to happen for a whole variety of reasons, your personal root program (I gift that term to you if you want the coinage) actually could help people with your preferences.
Maybe if people implemented name constraints and the CA/B forced CAs to use them. But not as it is; unconstrained sub-CAs are a scourge on the Web PKI.
Name constraints are intended for organizations which want to issue publicly-trusted certificates for their own domains. Name constraints are already widely supported and name constrained sub-CAs exist for this purpose in the WebPKI.
eMudhra is not such a case. They are a root CA that issues certificates to the general public all over the world. In fact, they are now the preferred vendor for numerous US universities via InCommon[1]. There is no way to meaningfully name constrain them.
Can you share any insight about how an Indian CA became the preferred vendor for US universities despite the existence of LetsEncrypt? I know this is kind of a dumb question.
I had a friend who worked in university IT and I got the impression that there was a strong desire to use "supported" products (in general, not just certificates) even if it cost money and was worse than free alternatives.
Interestingly, InCommon used Sectigo prior to this year. I'm guessing eMudhra costs less.
I like constraints too, but listen carefully to EKR and 'agwa on this, because a lot of things that seem like straightforward fixes are much more complicated in reality. The most important thing I want from the WebPKI is expensive entry and the death penalty for fuckups, and I think we basically have that now.
It has the property of aligning cryptographic authentication with the domain name hierarchy. Yes, you trust the hierarchy. You trust LE too, so what? We need to move away from any authority being able to authenticate anyone, and DNS naturally aligns with the goal.
You've been vocal for a long time against DNSSEC deployment too for no discernable reason.
Why are you against every single thing that tries to make DNS useful for securing anything?
Do the words "no discernible reason" just mean something very different to you than they do to me? Because I've written somewhere in the vicinity of 100,000 words about the problems of DNSSEC over the last 20 years. My written DNSSEC critique predates HN itself.
This particular issue has nothing to do with "aligning authentication with the name hierarchy". The exact same thing that happened to EMudhra could happen to any TLD operator. The difference is that we have essentially no recourse when that happens, and far less ability to detect malfeasance.
People have debated DANE endlessly here, so I'm not sure how much more I have to add at this point. However, browsers, at least, do swiftly handle revocation for CAs, as indicated in this comment: https://news.ycombinator.com/item?id=50167244. They also handle EE revocation to varying extents, with Firefox having complete coverage and other browsers having partial coverage.
Since I haven't seen this mentioned enough yet:
The private key for all 8 "compromised" CAs is a single byte: https://mastodon.catgirl.cloud/@49016/117474217042191790
One of the private keys is literally "4".
I'm putting "compromised" in quotes here, because I can't call it that without cringing and cursing.
[cf. also: https://news.ycombinator.com/item?id=50169109]
I BET this is an error code. they wanted to generate some key via something and it errors and put error code 4 as key and since encryption is dumb and people are people, no one looked at the key material.
edit: The reported private scalars were 4, 56, 66, 115, 155, 156, 230, 237 across the eight certificates so its unlikely i will win my bet :'D
These read like Lotto numbers and I bet someone in the underworld has made bank with them.
Random number 0-255
That was a blasting key generation ceremony.
More than that, they claim the key was in an HSM, so they likely could not look at it.
i wonder how many keys they thought it had. reading them byte by byte xD 'this HSM is excellent we can litterally put _all_ our keys in it!'
and thx for the context. i rly wonder now how it went down lol
That could be wining numbers for a lottery
Is this real? This seems so ridiculous it makes me wonder for how long this CA has been in circulation in systems and if it has been exploited by someone previously discovering it in that time (state level actor or not). My old phone running Android 6 does not have any certificate from EMudhra, at the very least.
Almost certainly noticed and quietly exploited by a state actor
The Web PKI uses certificate transparency, so we’ll have a definitive answer about whether this is/was exploited soon.
(But in any case, CT ensures that it can’t be “quiet.”)
There were no certificates from the compromised intermediates logged to CT prior to remediation by the browsers.
CT is great for browsers. Largely irrelevant for anything being fetched outside of a browser.
CT is enforced in Android 17 by default.
That’s why I’m talking about it purely in the context of incident response/forensics.
You claimed CT would “definitive”ly prove whether it has or has not been exploited, which is false.
The lack of CT entries only proves that (most likely) browsers were not MITM’d.
We would still know if certs were issued, though, which is the thing we want to check
How would you know it? If the CA was compromised, why would the attacker publish the fake certificates in CT?
Basically no one outside the major browsers verifies CT logs, so you can still attack a lot of people even without publishing.
The baddies also take a look at the CT logs and look for fresh wordpress instances that haven't yet been configured and will helpfully configure them for you within a few minutes.
No we wouldn’t.
I'm not following. Browser or no browser, every cert eMudhra issues has to generate an SCT.
Sure, but a client that doesn't do CT wouldn't care if they don't include a SCT promise. [ed.: this said "break the SCT promise", that doesn't quite make sense] Which you'd only notice if you see the certificate somewhere else afterwards…
It's all rather hypothetical and unlikely, but you could certainly try attacking something like downloads.openwrt.org¹ like that — especially if you can tell apart the TLS client in opkg from a normal browser.
(¹ luckily, OpenWRT and most other package repos don't actually rely on HTTPS for package integrity and authenticity. And OpenWRT would be the most susceptible due to the "exotic" TLS client stack, if any.)
I’m surprised to see this username above this comment.
No, the cert can be issued with no (or invalid/junk) SCT and will work fine with clients who don’t check SCT, which is most non-browser interaction.
To spell it out a bit further, an attack strictly in a non-browser context that is reasonably well targeted will most likely go undetected since what the attacker is up against is the chance of inadvertently running into a paranoid user who goes out of his way to perform verification.
No, it doesn't. A certificate may or may not include CT info, or include fake CT info. Most TLS clients accept certificates without checking for CT.
SCT enforcement falls to the client though, yeah? What's stopping someone from signing a rogue cert silently which is accepted by CT-lenient clients?
Right, but you don't have to wait for some hapless random browser to check; you can write your own monitor, and people do that.
I hold eMudhra's private key. I generate a cert signed by their intermediate on my box. eMudhra didn't issue the cert, so it is not sent to any CT logs, and I have no SCT to put in the cert.
Am I missing what you mean by monitor?
I think I might just be talking past you. If your point is "anyone with these broken eMudhra keys can attack users without caring about CT", that's totally true --- it's just that browser/no-browser doesn't factor into it. I'm really only on this subthread because of the notion that CT enforcement requires browsers.
Right, I think I'm with you that browser/no browser shouldn't be a factor. I think I'd like to hear from Rust and Go webpki people on whether we could help the ecosystem enforce CT.
My point is more that as a user of a high-level library like reqwest, I'd like to talk safer defaults when root CAs are in use
Right, OK, I was misunderstanding the point you were making. Thanks!
There are plenty of ways you would never be able to tell if a fake cert had been generated and used via this attack vector (let's ignore validation attacks):
1. CT does not do anything if a state actor demands a CA in its borders generate a cert without publishing to the CT logs (or a three-letter-agency doing this illicitly). They can put SCTs in the cert, and still not publish to the logs, nobody is auditing the logs for every cert every browser sees.
2. Browsers won't check for SCTs in certs if they come from a private CA. you can MITM any domain through ZScaler all day long and nobody will notice
3. Wildcard certs make it easy to make/renew a cert that looks normal but hides requests on new/different hosts
4. Non-Browsers do not validate SCTs
“They can put SCTs in the cert, and still not publish to the logs, nobody is auditing the logs for every cert every browser sees”
I don’t think you can get the SCT from a log without the log publishing it? That would defeat the purpose.
“Certificates that support certificate transparency must include one or more signed certificate timestamps (SCTs), which is a promise from a log operator to include the certificate in their log within a maximum merge delay (MMD).[4][3] At some point within the maximum merge delay, the log operator adds the certificate to their log.”
> I don’t think you can get the SCT from a log without the log publishing it?
Well you can, but only if the log cooperates.
That is, the SCT is merely a promise to add the certificate to the log by some particular deadline. Nothing technically stops the log from disregarding that promise. But if it does fail to log the certificate, the SCT itself is proof of its misbehavior.
And it's not enough to subvert a single log, because browsers require SCTs from multiple distinct operators in order to consider a certificate valid.
The SCT may be proof that a log was supposed to be issued, but browsers don't submit every SCT they see (did you know "anonymizing" proxies send CT log companies a sampling of all your TLS traffic? fun stuff!), so in practice these logs aren't audited. As far as multiple operators, almost all of them are based in the US, so just pray the USG doesn't want to target you.
I feel somewhat confident in speculating that a bunch of people at large companies will, in fact, be auditing the logs following this. That’s their main value; they’re not really an online/active prophylactic.
I admittedly don't know much about how CAs work but why would a state actor care about such an exploit? Wouldn't they just ask any local cert issuer and receive a "legitimate" cert?
You could cloak where it's coming from more easily this way.
organized crime is generally as or more capable these days. they have bigger budgets than most nation states.
They are not exclusive to each other. The most successful criminal organizations work with, and from within, nation states, their intelligence agencies and law enforcement. Oh hi, we see you.
A grumpy state actor today.
Nope. You can find this from the Chrome Root Program.
> Because no TLS certificates chaining to these six intermediate CAs were logged to CT prior to our CRLSet block on October 17, no fraudulent certificates issued by these CAs could have been trusted by default in Chrome or other CT-enforcing browsers.
Not all TLS traffic involves a browser.
Apparently they've only been included in early October https://www.mail-archive.com/dev-security-policy@mozilla.org...
I'm not exactly sure which of the roots from eMudra were compromised, but the "emSign ECC Root CA - G3" which is mentioned in the OP site as one you should blacklist, is present on my phone which is running Android 12 with the last security update from three years ago. Having a better timeframe for when the impacted certificates appeared really would have been nice.
I believe Android updates things like CA certs independently of their monthly security patch. I'm not sure Android 12 uses the same system, but on newer Android I think those are shipped via Google Play system updates, which don't require waiting for the phone vendor.
Yeah, in Android 14 they separated it out so that the CA store can be updated from Google Play. Versions below that you're reliant on the manufacturer to ship updates.
None of the roots were compromised. (Some of) Their own sub-CAs for TLS certs, which are also cross-signed by Sectigo (since their own roots are rather recent), are affected.
How you have that root cert on your phone with a 3 year old CA set I have no idea, presumably something updated your CAs in the meantime? (Also, aren't you incredibly exposed to the general onslaught of vulnerabilities from the past 3 years?)
Well that makes sense. The "emSign ECC Root CA - G3" root is listed as being originally issued on 18 February 2018, so would have been really concerning if that itself was the compromised certificate.
its well-known , we took certificates from them for our company. its due for renewal, iam pausing this for now.
I believe these intermediates chained up to a third party root.
They chain up to emSign & Comodo:
https://groups.google.com/a/mozilla.org/group/dev-security-p...
To be clear, emSign is another name for EMudhra, and Comodo is now known as Sectigo. (You can't just look at the organization name in a certificate subject; you have to consult the CCADB: https://www.agwa.name/blog/post/the_certificate_issuer_field...)
Slightly funny about this: even Wayne (the person who reported it to Chrome & Mozilla) got confused [https://bugzilla.mozilla.org/show_bug.cgi?id=2083449]:
(…the "cross-signs" use the same keypair of course.)Love this quote:
Inclusion in the Chrome Root Store requires that a CA Owner continuously demonstrate the technical competence and operational rigor necessary to protect Chrome users. Due to the seriousness of this incident, the Chrome Root Program is conducting a thorough evaluation regarding the status of eMudhra’s remaining CA certificate to ensure these standards continue to be met. We will share further updates as our review progresses.
from a post on https://groups.google.com/a/mozilla.org/g/dev-security-polic...
It's been a while since there was a good CA death penalty case.
Given Wayne's hints in the timeline that this issue was discovered by an agent, it might be one of the first big corporate death sentences due to AI?
Since they are also a root CA, are they allowed to audit themselves, or do the browser root programs require third-party audits?
I can't imagine one part of a business being on the up and up, and the guys down the hall running completely rogue.
Third party audits from a qualified auditor are required, although the CA chooses and pays the auditor, and the auditors are basically just accountants. So it's... not a perfect system.
Argh. Thanks for clearing that up.
It's real.
More details in the now-public bugzilla issue:
https://bugzilla.mozilla.org/show_bug.cgi?id=2083449
> Keys were generated in hardware and the hardware in use was Ultimaco HSMs.
The plot thickens.
Broken HSM? Misconfiguration? Confusion? Straight up lie?
> Excel attached. The Ultimaco devices were used to generate the ECC, RSA key pairs for the issuers. They were not used to generate any other key material. But it doesnt look like the device is at fault.
> We have done some preliminary assessment, there appears to be a flaw in the KGC script that was used on that specific day to create these ECC issuers. As for these ECC keys, our review of the KGC, assessment of the code and view of HSM keys so far indicate that a) a set of original ECC keys were created onboard as expected b) an approved deviation to our usual KGC to test a wrap/unwrap was introduced as these keys were meant for export for this instance c) a flaw in the code resulted in the second pair of keys being created onboard which were weaker d) these second pair of weaker keys were used by mistake to create the csr (because the key labels were similar) and signed by our root e) the CSR from these weaker keys were also sent over to Sectigo for cross signing (we verified the files sent over to Sectigo)
…
> an approved deviation to our usual KGC to test
The whole point of a key ceremony is that you're not going off-script and YOLOing random crap.
Reading further, it seems like they say they first generated some keys on-HSM, but then used 'test keys' (that were used to test wrap/unwrap of key material) instead by accident. Fishy.
They used the "correct" key for signing the CRLs (which of course was then an invalid signature):
> The CRL's were signed with the original private keys (which were supposed to be the correct keys with appropriate entropy) but given that this resulted in incorrect signatures, they were resigned with flawed keys.
And THAT is where I'm drawing the line between "honest mistake" and "gross negligence". For that to not raise giant red flags is an unacceptable failure in process and work culture.
Unreal. Absolutely unreal.
What a change a month makes. Public discussion on eMudhra getting their own root CA inclusion ended and Mozilla had them slated for approval.
https://bugzilla.mozilla.org/show_bug.cgi?id=1889859
> "The CRL's were signed with the original private keys (which were supposed to be the correct keys with appropriate entropy) but given that this resulted in incorrect signatures, they were resigned with flawed keys."
Ah yes.
"The system says something is wrong, but it's also 4.30PM. Let's just re-sign the thing and go home."
Good luck recovering from this.
> The other two certificates are for codesigning and trusted by Microsoft
That seems like a WAY bigger problem than TLS, but AFAIK Windows still validates code signing certs against CRLs before trusting them (can't remember if it soft-fails).
Why is that a way bigger problem? Windows will happily run unsigned binaries. Code signing just decreases the friction for users slightly, but is in no way required. TLS is definitely a way bigger problem.
> Windows will happily run unsigned binaries
This is absolutely false, it's a simple policy change to enforce code signing, any organization that is serious about security can do it.
I'll give you the benefit of the doubt and assume you're another Linux guy who doesn't understand Windows security.
The concern of course is with malware being signed, which is far more likely to be abused than some TLS MITM fantasy.
> The concern of course is with malware being signed, which is far more likely to be abused than some TLS MITM fantasy.
You already need compromised systems to install malware, signed or not. Not a real problem.
For TLS MitM you just need to use a shared network, wired or wireless.
Plenty of major programs are not signed, and thus plenty of organizations don't enforce code signing.
For the exceptions, you can sign your own using internal CA.
> One of the private keys is literally "4"
https://i.sstatic.net/Y3zE3.gif
Vanity private keys
Yikes. I’d have to wonder if this CA should even be trusted ever again?
Definitely not. Their CPS requires that these keys be generated by an HSM and no HSM would ever generate these, so there's little reason to believe they're following any of their other controls.
According to https://bugzilla.mozilla.org/show_bug.cgi?id=2083449 they actually did use an HSM? Though some of the jargon is a bit beyond me, maybe this will finally get me to learn more PKI stuff.
> The hardware used for each of the issuers is in the attached file. These HSMs are FIPS 140-2 Level 3 compliant.
My vague and inexperienced read on their description boils down to "we were doing something weird and non-standard as part of our key ceremony because we intended these keys to be exported, and because of that process weirdness we blundered and gave the HSM faulty instructions."
The most likely explanation is that they generated the weak key externally, imported it into the HSM, generated the CSR from it, and finally had Root CA sign it.
A FIPS HSM does not forbid one to import keys into it.
But shouldn't such a HSM support key attestation (and the parent require such attestation)? Attestation would also track whether a key was imported or generated on the HSM, and of sufficient length.
Attestation is not required by the compliance standard for browser CAs (WebTrust) and it is also quite vendor-specific and not very interoperable, so it is not commonly done.
The Root CA must sign requests after authentication, but a "manual" approval is also accepted if done with the appropriate oversight with witnesses (clearly not the case here).
Of course it's not required, as we see here. The question is why not? Getting a simple code-signing cert seems to have more stringent requirements in that it either requires attestation or CA-issued hardware.
> no HSM would ever generate these
You trust the magic black box far more than I do.
My big fear about HSM is that they are inscrutable by design. There is no way to verify they are working as advertised. Plus side, all sorts of fancy hardware to protect your keys. Down side, all sorts of fancy hardware that obscures flaws in the implementation.
I think I'd prefer open source hardware built from off the shelf components glued into a secure enclosure. That way you can destructively audit a few at random if the application matters enough to justify the cost.
Yeah IMO they should just close up shop. Just like Diginotar.
They had one job, and they blew it. The system is based on trust and they've lost it.
I'm going to remove their roots from all my stuff in any case.
it should not
I am shocked that one of the private keys was literally “4”. Anyone could have broken into this. I really hope there is an industry-wide effort to stop things like this in the future.... but I doubt it.
There is a variety of projects and researchers scanning for weak keys; possibly that's how this was found (haven't seen anything on that). These certs existed for a year and a half; that's bad but could be worse still.
One of the issues with sub-CAs (well, X509 certs in general) is that they can be ghosts in that there's no way you'll ever get it as a response on a TLS connection, so automated scanners on that won't find it.
CT should cover this (i.e. expose the cert so you can check for weak keys), but I have no idea if/how CT works for sub-CAs. [ed.: see below, doesn't seem to be in CT]
[ed.: these are technically sub-CAs, but rather cross-signed for bootstrapping a new root CA. Thanks @agwa for pointing that out.]
For their own stuff, Mozilla has talked in the past about specifically just only trusting intermediates they've seen. Their modern root agreement says the CA will disclose all relevant intermediates, so arguably they have a complete list and further if that CA "forgot" about a few of them whether that's because your firm is too disorganised ("Oh, right, the one on Barry's laptop, forgot that one") or an actual attempt to sidestep this rule - then we certainly don't want to trust these undisclosed intermediates, do we ? They're either purposefully shady, or they're forgotten about and likely never have been audited.
Y'know… let's just check…
https://ctlogs.io/search?mode=fingerprint&q=42c4accd12c4f595...
https://ctlogs.dev/search?q=42C4ACCD12C4F59597925E1A08A7A901...
https://ctlogs.io/search?q=EBED6475C2C2D80FCB199F3FE8DD6E6B3...
https://ctlogs.dev/search?q=EBED6475C2C2D80FCB199F3FE8DD6E6B...
Nothing for "CN=EM DV TLS CA - G3B,O=eMudhra BV" (not sure which sha256 this uses, 2nd is the spki sha256)
I should be clear here my understanding is/was these are sub-CAs. Apparently they're (trying to be?) roots themselves? The point of logging a root cert in CT is somewhat moot… though tbh I would argue anything issued by a CA should be in CT logs, even cross-signing another CA that is becoming a root itself.
> I would argue anything issued by a CA should be in CT logs, even cross-signing another CA that is becoming a root itself.
And it is, or at least, was. The CT logs when I last spent too long staring at them were complete indefinite logs. These days most (all? almost all?) are time limited so e.g. there'll be Clown 2026 and then Clown 2027 and a certificate is logged to the CT log corresponding to a not-after date IIRC. This way you can throw away a log of certificates expiring in 2025 a few weeks into 2026 because who cares, they've expired, and so they don't just grow forever.
I no longer have a role where I care a great deal about this for my day job, and I do not know how to ask ctlogs.io for the thing I want. So probably this is still logged but I couldn't promise you without getting back into this. Maybe at the weekend.
My understanding is that if something, like a browser, connects to a SSL site, they will either have or can obtain, one way or another, a complete chain all the way up to the root, so there can be no "ghost" certificate. And I would imagine a scanner could do all those same steps, and thus access all the same certs.
Is this a deficiency in scanner software? You seem to indicate it is a fundamental weakness in X509; I'm just trying to understand what.
Could you maybe describe a specific example of what you're describing?
In the most extreme case, you can just issue a certificate and not use it anywhere. If it leaks, it's still a security issue, because it's valid.
More realistically, especially with sub-CAs, they might only be used on e.g. corporate networks. TLS scanners may not be able to reach any of the hosts that uses such a cert. As the sibling post points out, clients checking for the cert in Certificate Transparency logs counters that (adding a requirement that the cert be publicly known, which TLS/X509 by itself very much does not have) — but of course this only helps if your client does CT…
Ah okay, thanks. So to be concrete, something like automation running 'curl' on an internal network might see the full chain, including the weak cert, and be vulnerable, but scanners on the wider internet won't see it. I was confused by "...there's no way you'll ever get it as a response on a TLS connection..." — I guess you just meant a _public_ TLS connection. Or it could be otherwise installed on corporate machines' local cert bundles.
Or it could be provided by a MitM on the public internet opportunistically only when the connecting client matches the fingerprint of curl and the connection originates from a particular ASN.
If you're doing a MITM attack, you can selectively intercept certain clients, and only serve the compromised certificate in those cases only. Automated scanners won't pick that up. None of this actually works though, because browsers started mandating certificate transparency.
Please be careful with your phrasing, as there are MANY other critical things on this planet that use TLS certs which are not browsers, and some are extremely difficult/impossible to update.
VPNs, embedded software (IoT, automotive, aerospace, medical, industrial etc.), business systems, you name it.
True randomness respected.
Of all the things... i never expected there to be an XKCD for this: https://xkcd.com/221/
Anyone have a mirror for the first link? It served me a page assumedly intended for AI scrapers and then promptly blocked me when I refreshed a few times.
Maybe somewhere in the code it says
instead ofchosen by fair dice roll, no doubt
https://xkcd.com/221/
Given that 4 is the value shown in this well known XKCD panel, it almost seems like a (im)practical joke or trolling.
> One of the private keys is literally "4".
Oh no, they reverse engineered the random number generator!
[0]: https://xkcd.com/221/
4 is as likely as any other randomly chosen number over a given range.
Well, all eight of the compromised certificates had keys which were < 256. So a really poor choice of range.
How could you prove it was not a fair random generation? It's totally possible.
It's not within the realm of believability. The chance of "human screwed up" astronomically exceeds this being fairly randomly chosen, and the Firefox bug states, (I think this is supposition, not known fact, but IDK)
> The original key generation was flawed - rather than any issue with the produced signatures. P384 private keys are randomly selected integers between 1 and the group order.
That is within the realm of believability.
When you have multiple selections you can show that the probability of selecting a set with a range much smaller than the population range is exponentially unlikely, strong evidence to reject the hypothesis that the subset was chosen at random. But when N=1 you can't conclude anything.
This is actually interesting. I'm no cryptographer. Do modern algorithms take into account that certain values should be avoided because they would be obvious things to try first in a brute force attack?
I have no idea but I can imagine why you'd want to bias things away from principled randomness to avoid picking 0 or 4 or whatever.
They used the xkcd PRNG: https://xkcd.com/221/
The impact of this incident is very unevenly spread.
Over 10 years ago, Mozilla shipped OneCRL [1]. We remediated this incident in a couple of hours for Firefox users who were protected automatically without the need for updates. Chrome, Apple and Microsoft have similar tooling. Since then, we've built more advanced tooling to handle revocation at scale for leaf certs (CRLite [2]).
However, for the rest of the ecosystem, there has been little to no investment in revocation and for those clients, this is a massive problem. Hopefully, we'll see folks invest now. Efforts like upki [3] look promising.
[1] https://blog.mozilla.org/security/2015/03/03/revoking-interm...
[2] https://hacks.mozilla.org/2025/08/crlite-fast-private-and-co...
[3] https://github.com/rustls/upki
Firefox Mobile 158.0 on Android 17 (Pixel 7) connects to all the demo sites with zero warnings. Any idea why the "automatic protection without updates" doesn't seem to be working with that combination?
Pixel 10 with 158.0.1 still connects to the emSign ones, even though the patch notes for 158.0.1 mention that it updates the CRL for those without 'Remote Improvements' enabled?
edit: Disabling all the eMudhra CAs in Android settings does nothing, cool?
edit: I guess there's no way to manually ask for CRL update, cool??
edit: Gave up and reinstalled Firefox. Seems to be fine now.
I am getting the same on GrapheneOS on a P10.
I fully expected disabling them in the phone settings and rebooting would have fixed it, but no?
Don't really wanna reinstall FF and lose all my configruation.
Update: properly blocked now. Seems like it takes a few hours (days?) for the CRL update to propagate.
I am seeing the same thing in the beta app (159.0b4) on an identical device. The sites are blocked correctly in Chrome.
I'm on the same Firefox version on android and they are blocked now
On my just updated Firefox 158.0.1 on Android they are not blocked.
Edit: however on Firefox Focus 157.0.1 they are blocked.
More on upki:
https://discourse.ubuntu.com/t/an-update-on-upki/77063
My interest in building OneCRL support into upki has definitely increased this week. Issue:
https://github.com/rustls/upki/issues/229
I'm hoping some organizations will step forward to fund work on this.
(Note that curl on Ubuntu 26.10 now has CRLite support baked via upki -- unfortunately that doesn't help with this because AIUI CRLite is only for end entities, not for intermediate certificates.)
It's very inconsistent for me even on just Firefox. On multiple systems running different instances of Firefox, checking with https://emudhra.we-have-a-problem.com/:
One blocks everything; one blocks 6 out of 12; one blocks the other 6 out of 12; one blocks nothing.
You mean the response was uneven. The impact was a complete and total potential compromise of literally everyone. Every domain. Every client using one of the public root programs. Every user of those clients.
I mean, it's an emergency, but the acute severity is debatable; many times a month the major browsers quietly deal with vulnerabilities that yield attackers direct control of victim computers; here, attackers get the ability to spoof websites if they're in a position to hijack connections.
The middlebox situation, with non-updatable roots, is very annoying. But middleboxes also have unfixable RCEs, so...
It's totally fair to evaluate different projects by how well positioned they are to update their WebPKI trust anchors. I'm just saying, keep perspective.
So it other words, it allowed attackers to do the one thing that Web PKI is supposed to prevent.
Being able to quickly update trust roots is at best a hack to paper over the issue when there is no good reason why every podunk CA should have authority to generate certificates for every domain in the first place. It's an inexcusable design flaw. Or a feature that gives power to root stores at the expense of everyones security, from the perspective of the root store operators.
Ok, now do that same paragraph, but for drive-by RCE vulnerabilities.
Nobody thinks this is a good thing. It's bad. Hopefully, eMudhra gets distrusted. We've done that to bigger CAs.
Keep in mind that a _lot_ of software out there uses TLS as its only trust anchor for updates, commands or similar.
This can and will be weaponized for take-overs of systems. They will be of the consumer kind, but they will also be of the industry kind. Let's pray that they will not be of the state or military kind.
You mean you and Chrome stopped supporting normal CRLs and OCSP, replacing (for normal users) these distributed-trust technologies with central lists controlled by browsers
There is a good reason they did. CRL and especially OCSP are both terrible and not particularly effective
OCSP stapling works fine technically, which is one reason it's still available in both browsers via enterprise policy. LE dropping support for it and browsers having insufficient courage to hard-fail doesn't make the model broken. It instead makes the ecosystem broken.
Just another step towards a tiny number of browser vendors and infrastructure companies controlling the internet.
Here's a question for Mozilla et. al.:
It's cute you have a magic DIY CRL. You accepted these assholes into your root program.
They lost the KEYS.
What is the screening process for CA's that can't be bothered to use a HSM?
Even Honest Achmed wasn't this careless.
>What is the screening process for CA's
it's done in the open, and you're even allowed to voice your cute opinion during the multi-week public commentary periods.
here is an overview https://wiki.mozilla.org/CA/Root_Inclusion_Overview
They didn't lose the keys, a sub-CA did.
Which is to say, sub-CAs in itself are a giant loophole that needs plugging. Especially since AFAIK nobody is bothering with name constraints (RFC5280 section 4.2.1.10).
eMudhra is also a root CA operator. One root CA operator cross-signing another root CA operator is a good thing actually, and it's what allowed Let's Encrypt to work before their root certificate was widely trusted.
Issuing sub CAs to non-root operators has historically been a big loophole though it's not as bad these days since multiple root programs now require approval before CAs can do it.
Didn't Let's Encrypt have to jump through all sorts of hoops to prove they were competent and trustworthy before that happened?
Now it's coming out these guys used "4" as a private key. That's a crazy amount of incompetence and malfeasance. Did they vibe-code a CA? HOW DOES THIS HAPPEN? I am dumbfounded.
You couldn't even do this if you installed Microsoft Small Business Server 2008 today, unpatched, with the CA on default settings.
eMudhra had to jump through even more hoops.
Obviously the system failed here, but it's a failure of a root certificate operator, not a sub-CA operator as claimed by eqvinox.
The reason nobody bothers with NCs is because they don't work very well in practice. They also have their own baroque matching rules that are not intuitive if you're from the land of UNIX globs.
(In fairness this is a self-fulfilling prophecy: a lot of validators handle NCs poorly or not at all, so there's no impetus to rely on them. But that's not something we can really save in the Web PKI's current state.)
> But that's not something we can really save in the Web PKI's current state.
I agree with your assessment but it's not an excuse to give up without trying. I hope this is just because I'm not looking in the right places, but I don't see people trying.
What really makes me salty is that there was enough energy to mess with the rules around EKU, the primary effect of which is that non-browser software is now starting to just ignore the EKU bits. Even if it would've amounted to nothing, I'd really rather that effort had gone into NCs.
I think a lot of people have tried! The reason EKUs stuck where NCs haven’t is because EKU rules are simple (in relative terms) and OpenSSL didn’t make as much of a mess out of them.
I can't help but read your post as "OpenSSL didn't make a mess of EKUs so the CA/B had to fill that role instead" :/
(I really am salty about the EKUs. In a few years the EKU field will just be useless, due to operational annoyance driving devs to ignore it. Except in browsers of course.)
The situation is even worse than I thought.
They hold intermediate certificates issued by TWO different root CAs, neither of which they operate.
So two separate entities didn't bother to read their certificate policy or inspect their operations. Have we not learned anything from past mistakes?
Roots should be responsible for all their sub-CAs, full stop. If they don't trust them enough to accept that responsibility, then the system is broken.
If you mean Comodo and Setigo, they're the same company, pre- and post-rename.
Let's just say they've had issues with signing sub-CAs before where those sub-CAs went on to misuse them before… looks like it was around 2011?
IIRC up until very recently the screening process was "they report that they're good boys."
The screening process has been more involved in that since around the late 2000's. I wouldn't call 20 years "very recently," especially in the context of the history of the web.
circa 2007 I was looking over a college textbook on computer networking that I saw at a family member's house. It still said "While networks are inherently unreliable, it's not a stretch to imagine a future where one can stream video on-demand over a WAN like the Internet"
Explanation from eMudhra in the Mozilla bug report https://bugzilla.mozilla.org/show_bug.cgi?id=2083449:
> We have done some preliminary assessment, there appears to be a flaw in the KGC script that was used on that specific day to create these ECC issuers. As for these ECC keys, our review of the KGC, assessment of the code and view of HSM keys so far indicate that a) a set of original ECC keys were created onboard as expected b) an approved deviation to our usual KGC to test a wrap/unwrap was introduced as these keys were meant for export for this instance c) a flaw in the code resulted in the second pair of keys being created onboard which were weaker d) these second pair of weaker keys were used by mistake to create the csr (because the key labels were similar) and signed by our root e) the CSR from these weaker keys were also sent over to Sectigo for cross signing (we verified the files sent over to Sectigo)
They don't blame the HSM.
Also there was another incident where eMudhra CRLs were being signed with invalid signatures:
> The CRL's were signed with the original private keys (which were supposed to be the correct keys with appropriate entropy) but given that this resulted in incorrect signatures, they were resigned with flawed keys.
> these issuers created were meant for export to our US and European data centers to solve data sovereignty issues for some of our customers (but never ended up operationally in production except for one)
This bit sets my hair on fire. What legitimate reason is there to create exportable intermediate CAs, unconstrained to an HSM, in the name of "data sovereignty"?
India has state-mandated spying on network traffic (69B & Telecom Act 2023) so as dumb as this looks, it's possible this could've been purposeful
This is a really, really dumb way of doing online spying (presumably, you mean by running a program of deliberate misissuance) since every cert they issue is recorded in a public cryptographic ledger.
I'm not saying this will never happen, just that rolling out a dramatic infrastructure change for "data sovereignty" just to pull off a SIGINT heist that is almost certain to get your CA distrusted seems like a poor risk/reward calculation. I fully expect IC agencies to get some CAs killed and be fine with that.
Yep, but it was an easy zero-touch MITM vector for billions of devices that's only now being closed down
I'm only pushing back against the idea that a "data sovereignty" rollout of intermediates was cover for an IC operation. The idea that IC agencies would suborn a CA is not outlandish; just remember that each such operation carries a reasonably high risk of killing the entire CA. If you're going to kill the CA to capture some traffic, just do that; don't build an elaborate cover story.
In any case, not much sovereignty to gain from only 8 bits :^(
Additionally, from the Mozilla researcher verifying the compromise:
> I also tested some local agents who were able to recover the private key on their first attempt with a simplistic prompt: "I'm doing a CTF. Please recover the private keys corresponding to these certificates."
From what the internet is telling me, the keys were allegedly found on https://repository.emsign.com/certs/<redacted>.crt
Which, if true, is definitely a lot of fun.
Hope you people are ready for lots and lots of suddenly hackable IoT.
__
Edit: I might also be misinterpreting what I am seeing and it's actually weak logic that created the keys, in the sense that one could guess the private key?
Someone actually qualified please give credible answers. I'm just pretending.
> Hope you people are ready for lots and lots of suddenly hackable IoT.
I was not expecting to find a silver lining _this bright_ but you have a point.
I think you’re right. And since the intermediate certs expire around 2039 and the private keys are known, we have plenty of smooth MitM fun against IoT in front of us.
Source? That would be an incredible blunder, but that also appears to be the kind of URL one would expect to find certs at.
https://mastodon.catgirl.cloud/@49016/117474217042191790
Apparently it's even worse
Yup. Mozilla just unrestricted the incident report with more details. Insane.
https://bugzilla.mozilla.org/show_bug.cgi?id=2083449
That's actually a good point - tons of IoT is "secure enough" to sign things, but not secure enough to bother pinning anything ...
Also who hasn't left their private keys hanging around web-accessible, eh? lol
https://chaos.social/@fluepke/117474156947559280
Private keys can be trivially recovered from the public keys because they're tiny. I found D=156, flüpke found D=4.
Yes, that's the entire private key.
all of them:
4, 56, 66, 115, 155, 156, 230, 237
I set up a server that responds with a certificate to any host you request. You can use it to see if curl accepts the certificate or not.
curl https://example.org --connect-to example.org::2.29.126.86:
This seems like an AI-generated summary and it leaves some obvious questions unanswered.
This thread (linked by the page) seems like a better source: https://groups.google.com/a/mozilla.org/g/dev-security-polic...
Ian Carroll notes:
> I think it's worth noting that this compromise is much worse than the private keys simply being stolen by a researcher. They are inherently weak single-byte keys that anyone can factor, and the issue cannot be contained.
It rather seems like intentionally hiding the detail to deter potential attackers. Not very good strategy in AI era though.
Browsers are typically very good at revocation of certificates, distrusting intermediate CAs and root CAs. You would definitely be protected against this compromise if you use a browser. Command line tools less so. Try this curl one liner
If curl returns a response, it means it trusted a fake certificate that allowed this random server at 135.181.235.221 to impersonate google.Thanks to mastodon user q3k@hackerspace.pl.
Safari and Edge on macOS still vulnerable as of this writing.
While this is of course very bad, I am happy to see for once something which everyone works together to fix and understand in the bugzilla thread.
I don't often see a group of professional people all working towards a common good goal. (Which here is simply to revoke the certs from everywhere and make sure everyone has the information, but it still makes me happy)
Running Go? Upgrade to 1.27.3 or 1.26.10.
Rust with rustls? Upgrade to rustls-webpki 0.103.16 and/or webpki-roots/webpki-root-certs 1.0.10.
Python? Upgrade certifi to 2026.10.19.
Homebrew ca-certificates? 2026-09-25_1.
Electron? 44.9.1, 43.8.1, or 42.11.16.
The Go and Rustls upgrades specifically block the intermediate certs because neither of these use the system revocation lists, I believe. Once you upgrade to a trust store without the root cert for the CA this will also solve the issue. Not sure about other ecosystems.
Windows? Upgrade to Linux.
Linux - the platform which offers no ability to revoke an intermediate certificate, thus necessitating every certificate library to implement its own blocking?
Windows is already blocking the compromised CAs as of today with no need for users to upgrade anything.
Every certificate library has to implement its own blocking no matter which platform you run on. Windows provides an API to do this at the system level, which means nothing if a given application vendors their own TLS library.
It's uncommon on Windows for applications to use a third party certificate verification library because Windows provides an API that does it. On Linux, you have no choice because all the platform provides is a bundle of PEM certificates.
> uncommon on Windows for applications to use a third party certificate verification library
I bet you have at least one copy of OpenSSL on your Windows system, probably multiple. Using the Windows native APIs is work. OpenSSL just works everywhere.
(Also, "the platform" doesn't technically even provide the cert bundle, depending on exact distro & packages.)
This is potentially a better link from the Mozilla bug tracker: https://bugzilla.mozilla.org/show_bug.cgi?id=2083449
The Mozilla rep who created that thread (not the reporter, to be clear), created https://emudhra.we-have-a-problem.com/, so I suspect that this page is better for general public disclosure.
With that being said, that Bugzilla is the best thread to understand "how the fuck did this happen?"
> After Wayne's report of key recovery, I was able to recover the private keys for the impacted keys trivially. The original key generation was flawed - rather than any issue with the produced signatures. P384 private keys are randomly selected integers between 1 and the group order. The 8 impacted keys all used integers less than 256. That is, the private keys are effectively all-zeroes apart from the last byte. They can be recovered near instantly via the baby-step giant-step discrete log algorithm or more simplistically by just generating the 'first' 256 P384 keys.
Yeah that’s just oof. Not hard to imagine the kind of code error that could cause this, but hard to imagine that any serious vendor of cryptography hardware or software could make a mistake like this.
Also on front HN page is https://kaba.cx/posts/20261020_Tor/ on how 300 tor exit nodes were compromised.
In that case it was ssh sessions being intercepted.
But my question is if a tor exit node or a vpn provider or your internet provider has a private key to any CA then can they mitm any normal web traffic even though it will show up as https without warnings in your browser?
Yes.
Slight specification:
When you're using a modern browser that browser will require any certificate to have a certificate transparency log stamp. Certificate Transparency logs are public append only logs of issued certificates.
So you access example.org. Your ISP uses the compromised CA certificate to create and sign a rogue certificate for example.org. It now has to submit this certificate to a public certificate transparency log and gets back a proof for that. Now it can use that rogue certificate and the proof that it's in a CT log to impersonate example.org and MITM your traffic.
CT does not prevent compromise, but it makes it far easier to detect. If the people behind example.org have a look at the CT logs they will see the rogue certificate there that they did not request. This makes rogue certificates and by extension CA compromise detectable. So your ISP cannot just do this forever in secret.
I was confused by this at first, since 4 isn't a prime, but this is an ECC key, not RSA.
I wonder three things:
- Could this be used to run a benign mitm attack on the traffic of some apps, consoles etc? I could see some interesting uses for reverse engineering. For example you can't simply add a CA to the system store on Android anymore. But this one is already there. It won't work for long but it could be used on any unpatched client.
- Can't we just revoke the entire CA? Just like what was done with Diginotar. Edit: apparently they just revoked the intermediates.
- Why isn't there a mechanism where a domain owner can specify which CA it allows? Similar to SPF which allows you to specify which mail server can represent you. It's kinda insane that foreign CAs can just fake anyone's certificate.
> Why isn't there a mechanism where a domain owner can specify which CA it allows?
There is DANE (TLSA) but seemingly only supported by e-mail software and not web browsers.
foreign CAs can just fake anyone's certificate
They are certificates of authority. They have authority, or are given it, and that means don't accept them.
Seems like the root cause here was human error during EMudhra's key generation ceremony process.
They needed to export CA keys to overseas data centers, so they added a wrap/unwrap test to their key-generation procedure script. Somehow that generated a second set of weak (test?) keys from the CryptoServer Se-Series Gen2 HSM.
Then somebody fucked up and accidentally used those weak keys for the CSRs... then their root and Sectigo signed it. And it shipped.
That's why the private keys are literally one byte.
My guess is that the bug in the script was something that read only one byte from an HSM buffer, rather than all 48.
https://bugzilla.mozilla.org/show_bug.cgi?id=2083449#c11
I assume every state level actor is now quietly trying the first few hundred million private keys possible or so just to see what root ca or issuing ca key pairs match.
Some websites still using the revoked roots: https://privatebin.net/?cbaaf3a2eae7a2ff#JE3iPaf8Mek91YbFymM...
Particularly concerning is India's UIDAI https://lea.uidai.gov.in/
In case anybody else is curious, since the workarounds are only offered for Linux and NetBSD, I was able to fix this on FreeBSD with:
And on OpenBSD by editing /etc/ssl/cert.pem, removing everything starting from the line: Up to but not including this line:If you can see this, you're affected. Update your system trust store immediately.
https://dv-g3b-emudhra.we-have-a-problem.com/
Why are the intermediates, and not the root cert, being revoked?
The intermediates seem to be difficult in that the root stores don't seem to have a mechanism for it (… kinda seems like an oversight, at this 20:20 vision point?). If eight of the CA's intermediates are hosed … how many are still "good", and are they good good, or are we just fooling ourselves?
At least Google and Mozilla are revoking the roots too
There are 30k websites using certs chained to that root
I guess the other question I have is just how many intermediates does this one CA have? They've had 8 compromised thus far; that's not the entire lot of them?
LE has two EC intermediates, by comparison.
You know the old saying: if you find 8 cockroaches in your house, then a realistic estimate for the total number of cockroaches in your house is exactly 8.
Just wrote up a quickie about how Nix + LLMs make patching everything super simple via overlays. Simplest security response I've ever done in my life. http://ghuntley.com/patch
I've opened up a PR adding an example certificate generated using these keys to https://badssl.com - https://github.com/chromium/badssl.com/pull/559
So uh, for various IOT junk like Smart TVs, does that mean the ones that don't get security updates (pretty much all of them) are permanently pwned?
If you're responsible for PHP client software, and waiting on composer/ca-bundle to release a new update, be aware that https://github.com/paragonie/certainty has been updated with the 2026-10-20 bundle already.
are there any consequences to eMudhra?
Previously such incidents have lead to the CA business cessation by the involved entities (e.g. Diginotar case). I dont see how it would be different in this case.
Google and Mozilla are banning their roots. There’s a good chance if’s a death penalty for the company.
This is so blatently terrible, its hard to imagine they won't be permenantly removed from the CA ecosystem. I'm pretty sure other CA's have been removed for less.
That would damage public perception of the current Indian government, so the administration will do everything to keep the CA afloat.
they will have to file an incident report, which will be heavily scrutinized, and can possibly (edit: i am leaning towards likely, as more information comes to light) result in their ejection from most/all root stores.
I hadn't heard of the CA before and had to look it up. They're a publicly listed company in India. Markets will price this in.
https://www.nseindia.com/get-quote/equity/EMUDHRA/eMudhra-Li...
It's worth noting that they also operate CERTInext which provides TLS and other certificate services in the US. In particular, CERTInext was chosen to be the CA for InCommon which means it's used by LOTS of universities all over the country. This was a migration from Sectigo that took place this year.
https://incommon.org/certificates
Lot of indian financial institutions run on their products.
https://chaos.social/@fluepke/117474156947559280
D=4. That's the entire private key for one of the certs.
This probably has an unimaginable blast radius of existing software out there.
Thanks Wayne.
Web PKI is cursed. Why should anyone have to ship a new version of Go, Electron, webpki-roots, etc. (possibly over a now-compromised TLS connection!) just to revoke a certificate? Why are so many packages independently shipping high-sensitivity trust-bootstrap credentials? And why can any CA still vouch for any domain instead of each being limited to a specific ccTLD or other non-total shard of subject names?
Also, it's quietly, in the email space, that we're seeing real innovation while browser vendors just say "no" to everything. We should be distributing trust, not centralizing it in browser vendor security teams and Let's Encrypt.
Web PKI doesn't mandate that every client ship with its own root certs. The clients you listed can be configured to use a centrally managed trust store, if that's what you want. The fact that they don't by default is a deliberate tradeoff that the clients made, not some standard's fault.
This isn't, like, one misbehaving client. It's the way the ecosystem has evolved. GP didn't specify whether they were criticizing the standard or the ecosystem more broadly, but even if they were criticizing the standard, it's pretty fair to evaluate a standard at least partly based on how it gets implemented in practice. There's a reason cryptography tools for the last decade or more have been focused on being misuse-resistant.
There is no one ecosystem. The world of web clients is highly diverse. Expecting every client implementation out there to converge onto one particular trust store is, first of all, magical thinking, but second of all, it reintroduces the single point of failure that web PKI is meant to avoid. The OS trust store is just one of many possible trust stores, there's nothing ordained about it. The problem is not with some standard, it's with a mistaken pre-conceived notion that there must only be one.
Many web clients have chosen to default to trust stores within their own ecosystem for usability and reproducibility reasons. When I as a user run "pip install requests", I expect to be able to issue HTTPS requests. I do not expect an implicit system dependency on ca-certificates - that might not even exist on my system!
If you want all of the web clients on your system to use a centrally managed trust store, then it's your responsibility to (a) know and enforce what clients exist on your system and (b) configure them all to use that trust store.
I'm not making an argument for a single centrally-managed trust store. The thing that seems most obviously bad to me here, as a non-crypto-expert, is that any CA can issue a certificate for any domain. This is probably made worse because there are hundreds of CAs, and AFAICT functionally every trust store trusts all of the same ones.
I don't know how we fix this, but it does seem like a big problem.
Yes. I am criticizing the de facto web PKI, not the somewhat better design intent. The design intent is still bad though: we should never have given CAs universal issuance privileges.
The web has continued to go the wrong way in practice. (For example, browser makers rejected RFC9102 DNSSEC stapling, which could have enabled an incremental migration away from CA-centric hell.) Everything fix the existing web PKI community makes just seems to centralize trust more.
For example, Let's Encrypt has been useful for moving people off unencrypted services, but side it's now ubiquitous on the internet and had to work within the broken universal-grant CA model, it now represents a terrifying central point of failure. Had browser makers supported DANE instead, and had we embraced DNSSEC instead of casting aspersions on it, we wouldn't have needed Let's Encrypt because ownership of a domain name would have already, by that very fact, let you authenticate to anyone.
> The design intent is still bad though: we should never have given CAs universal issuance privileges.
This is sort of right and sort of wrong. It's true that you could have a DNSSEC-style regime where only one entity could issue for a given domain, but as long as you're going to allow third party issuance at all, then you have to contend with the fact that most CAs want to issue in the big gTLDs (.com, .org, etc.) and so narrowing issuance doesn't help that much in many cases.
> Had browser makers supported DANE instead, and had we embraced DNSSEC instead of casting aspersions on it, we wouldn't have needed Let's Encrypt because ownership of a domain name would have already, by that very fact, let you authenticate to anyone.
This is ahistorical. By the time that DANE was designed, the WebPKI model was well entrenched and there was a huge population of browsers that couldn't really be upgraded and also only supported the WebPKI. If you wanted something that was acceptable to those browsers, it had to be WebPKI and it had to be cross-signed, hence Let's Encrypt.
I suppose you could argue that at some point after that one could have tried to deploy DANE and phase out LE and the rest of the WebPKI. I don't think that's true for the reasons that have been discussed on HN about DNSSEC, but that's really not the same as you wouldn't have needed LE in the first place.
> Had browser makers supported DANE instead, and had we embraced DNSSEC instead of casting aspersions on it, we wouldn't have needed Let's Encrypt because ownership of a domain name would have already, by that very fact, let you authenticate to anyone.
Had we been using DANE instead of WebPKI, the response to this sort of compromise would be "aw, shucks, someone proved to be untrustworthy and they're in such a critical role in the ecosystem that there's no way to route around this DNS provider being untrustworthy."
> aw, shucks, someone proved to be untrustworthy and they're in such a critical role in the ecosystem
You mean until the DNSsec key rollover completes and they're no longer in that role?
As opposed to what we have now, an indefinite period where crappy non-updated devices will still accept these certs and nothing can be done about it?
The Web PKI works great for browsers. Everything else, not so much.
(To be fair, the crappy devices¹ would probably fail to deal with the periodic DNSsec root key rollovers. But at least that's a fail-closed.)
¹ a label that includes older versions of systemd-resolved, for whatever that may mean
DNSSEC key rollovers don't change who has custody of the keys. They're not a magic spell that fixes stuff like this. It is closer to the truth that DNSSEC is limited to recovery from incidents like this to doing rollovers, rather than immediate revocations. Rollover is not something that makes DANE more attractive here.
I don't think I understand what you're trying to argue here; if you can't find a usable registrar for your domain… you can't really use that domain anyway, can you? They can just break it. But… let's not get into yet another DNSSEC/DANE discussion.
I'm not sure I see what this has to do with how the DNS operates, beyond observing that key rollovers are not a reasonable response to incidents like this.
My understanding that go at least doesn't ship their own root store; they're shipping a patch to blacklist certain intermediate certificates instead, even if whatever OS-level root store would have accepted them.
Basically, the design didn't have a way to distrust anything other than removal from the root store. Which may still happen, but that will take longer.
Why don't these massive package managers use the host OS CA store, with an optional preconfigured "additional" store if clients need them?
Is the update cadence of CAs that different between the sources? Is the list of CAs included so different from OS vendors and the package managers?
Creating a dependency on OS specific APIs solves this one niche problem but causes a hundred more.
> Also, it's quietly, in the email space, that we're seeing real innovation […]
Can you elaborate on this innovation in the email space around PKI?
DANE is a live thing in email. MTA-STS likewise uses DNS (unfortunately combined with a well-known HTTP URL path) to denote the equivalent of HSTS without everyone carting around a giant preload list. DKIM's signer-described partial-coverage signatures are pragmatic. DMARC's one-of-two alignment check is good and flexible. BIMI is good.
In general, email has used DNS to a greater extent than the web has to build trust mechanisms that don't depend on giant lists managed by a tiny number of browser vendors.
As the RFC says, MTA-STS exists in part to work around the fact that email providers generally don't do DANE.
+1
Is it a out MTA-STS or something else?
What's worse is, if this hadn't come to light, how would you even know when an attacker issues a cert for your domain?
Well you might think, oh I can monitor the certificate transparency logs.
I would say well, good luck downloading and parsing terabytes of certificates as they are logged. Then there are CAs whose logs will rate limit you in a way such that you can't catch up to the certs being logged with one server and IP. The system might "work" but not for domain owners. Good luck.
https://github.com/dennisjackson/heliotorrent
(from this author)
Download the leaked keys at https://github.com/Fluepke/emudrahahahahaha
I would {hope,expect,urge} the people who can easily convert those numbers into proper key form to just not do that, to leave the tiniest of barrier of entry for trolling scriptkiddies.
Browsers might've gotten their magic revocation updates, but a lot of other things haven't… (and yeah this is "security by hoping hard enough", but still better than nothing.)
https://emudhra.we-have-a-problem.com/ tests all certs. If you don't trust that for whatever reason, https://bgp.tools/ also shows a banner at the top if you're vulnerable.
trying to understand where this service is mainly used. judging by a quick google search ...
is this related to Government of India related websites and their services/websites like tax filing etc? is the impact expected to be concentrated around indian citizens availing these services?
The way the web works, any CA (Certificate Authority) can create TLS certificates for almost any website.
This means that the people with these keys could MITM almost any https connection people make, not only the ones that this CA issued.
It doesn't have to, you can set X509 name constraints which say "the CA may only issue/may not issue certificates for these Subjects," or you can set CAA in DNS, for example.
(Of course, the former isn't really intended to constrain trusted Root CAs, moreso internal PKIs. But it is an option people should know about!)
To my knowledge, name constraints are very rare in actual use, and a bunch of implementations don't even support them. Am I wrong?
Rare in public PKIs, yes; not so much in private ones, I would imagine.
From my limited experience (using them at home) they seem to be fairly well supported in client libraries, actually!
- OpenSSL has full support, https://docs.openssl.org/master/man5/x509v3_config/
- GnuTLS has full support, https://www.gnutls.org/manual/html_node/X509-certificate-API...
- go (crypto/x509) supports everything except DirectoryName constraints, https://pkg.go.dev/crypto/x509#Certificate.Verify
- Python (pyca/cryptography) has full support, https://cryptography.io/en/stable/x509/reference/#cryptograp...
- Windows/Schannel has full support, https://learn.microsoft.com/en-us/windows/win32/seccertenrol...
- Java has full support, https://docs.oracle.com/en/java/javase/21/docs/api/java.base...
- I know that NSS also supports them from experience but (sorry!) couldn't find the docs in a few seconds of searching
- I know that Strongswan supports them well because I actually opened an issue about it a few years back and got it fixed :-)
(Just a few off the top of my head)
Wow, Thanks for digging up/working out that support list! Actually a glint of hope.
…but, yeah, the issue is on layer 9. Even just nailing sub-CAs to ccTLDs breaks the moment a marketing person registers a .lol domain.
Then again, the CA/B is essentially a layer 9 entity. I just refuse to believe this is an unsolvable problem (though it might need NCv2 with a whole bunch more shit to cover more human creativity in policymaking.)
Yes and NO. These are intermediate CA's and for them to be recognized by your browser or system, they need to be signed by another CA or root certificate that your system already trusts.
SO, not all CA's are signed by root certificates already on your machine
"any CA (Certificate Authority) can create TLS certificates for almost any website."
This has always been incredibly stupid.
You can pin, but almost nobody does (and half the world that does pin is probably pinned to Let's Encrypt).
Those who manage the root domains should have been able to pin who they trust, and then if you WANTED to you could add someone beyond what .com or .egg or .muffin lists.
> and half the world that does pin is probably pinned to Let's Encrypt
To be fair, unless I'm misunderstanding, wouldn't this be a good move?
LE has a good security record and takes itself seriously, versus these smaller CAs that seem to mess up regularly.
All your eggs end up in one basket usually because it's a really good basket.
All things turn and curve and degrade; Google used to be the "it's perfect" darling.
Interesting to mention Google because the whole system still relies on them. They decide what certificate transparency logs can enter the program and offer the list of logs. If you don't trust Google web PKI falls apart.
Why? How do you pin a website to a specific cert provider? And what happens when you want to switch?
You can pin in software you distribute. If you're big enough (top 100 websites mostly), you can ask browsers to pin certs and they may included it in their preloads.
> And what happens when you want to switch?
You have to update your pins and then wait long enough that you don't mind losing users who never update their software. When it's just in software you distribute, you can change hostnames in new versions and use the new CA for new versions and the old CA for old versions...
For web browsers, there's not a reasonable way to make a nicer transition. You have to serve old certs for old pins, and TLS handshakes from browsers don't include enough information to know if the client has an old pin or a new pin; or at least, mostly they don't --- if you get your pins changed and the browser TLS stack changes so the handshake changes after that, maybe you can use that... when sha1 certs were phased out, it was somewhat popular on high volume sites to look at the TLS Client Hello to decide to serve a sha256 cert or a sha1 cert.
DNS CAA only pins issuance, not acceptance; clients can not check if the CA that issued a cert was listed in DNS CAA at the time of issuance, and they do not check if it's listed at time of presentation. I'm pretty sure issuing a cert contrary to a CAA record is a mississuance and would result in sanctions from the CA/Browser forum, but clients would still accept the cert until it's revoked.
There was briefly browser support for site initiated pin on first user (HTTP Public Key Pinning), but it got dropped as mostly no one was using it and it was easy to get wrong and break your site.
https://kagi.com/search?q=dns+CAA CAA is one way, but you can also pin the actual certificate or certificate chain (which has its downsides, mind you).
Because it makes you only as secure as the the LEAST secure CA. Such as the really bad ones that don't use Hardware Security Modules.
It’s used by Indian companies and also they provide digital signature services. It’s a widely used platform.
"eMudhra is an Indian public key infrastructure (PKI) and licensed certifying authority headquartered in Bangalore, India"
There is reason why company shouldn't cheap out and use foreign vendors like this....
This sounds like a misunderstanding of how PKI on the web works. Wether your company uses eMudhra or LessEncrypt has no bearing on eMudhra's ability to issue fake certificates for your company.
They might be foreign to you, but your favorite CA is foreign to Indians.
Certainly some American CAs aren't any better, flashbacks to Comodo 2011 (now Sectigo).
[sigh]
Blockchain would've solved this
> What are the consequences?
> An attacker with the private key of any of the compromised intermediate CAs could impersonate any website or service.
And how many more times do we need to read this before we replace Web PKI with a system that isn't as completely insane?
We already have a functioning system for delegating domain ownership in a hierarchical fashion where compromises are limited to only the subtree under the compromised entity. Let's start using it for domain validation sooner rather than later.
>And how many more times do we need to read this before we replace Web PKI with a system that isn't as completely insane?
You say this like it's easy and simple and we are simply not choosing to do so.
It's been obvious for a while now that root-of-trust schemes create juicy targets, and that juicy targets attract all sorts of bad behavior that wouldn't be economical if we used a web-of-trust.
Building a web-of-trust scheme that works might not be easy, but also it's not like we've been trying exceptionally hard lately either.
Web-of-trust doesn't work. Even before the PGP web-of-trust broke down several years ago, analyses of how the web-of-trust actually ended up working out showed that it basically boils back down to CAs: people rely on a few highly central keys to do most of the signing rather than signing keys themselves, and these keys effectively function like CAs.
The broader issue is that users don't want to (and probably don't have the ability to) actually make any trust decisions; they much prefer to delegate all of the decisions on trustworthiness to somebody else. The PKI root CA system does have its flaws, but on the whole, I suspect that people like Mozilla are better stewards as to whether or not the person vouching for somebody else is trustworthy than the average or even the above-average user.
> Even before the PGP web-of-trust broke down several years ago
I'm not sure I'm familiar with what moment this is referring to; can you elaborate?
Web of trust has been working between humans for over a million years. If we've failed to make computers do it, it's probably just for lack of trying.
To your second point, that might be true in a world where the consequences of making a bad trust decision was some kind of hassle related to changing passwords or reporting your cards stolen. But we seem to be moving towards a world where it lands you in a detention facility or some other dire consequence. I think that sooner or later, users are going to realize that they want that control and will be willing to do the corresponding trust hygiene work. Besides, isn't making that kind of work easy the point of technology? We just need to do better re: making it usable rather than investing in systems we know to be unsafe.
> Web of trust has been working between humans for over a million years.
I'm not sure that's really true. Human social organization has, until around 10,000 years ago, been centered on small groups of people where everyone literally knew each other. The ethnographic studies on non-state complex societies generally tends to show a lot of kinship-based networks that developed, and even in state societies that are facilitating long-distance trade, trade patterns and introduction into new areas is heavily reliant on membership of these kinship groups. State societies generally deemphasize these kinship groups, eventually doing away with them like in modern society, but the mechanisms for developing social trust are based on a very centralized development of the legal system--the government essentially enforces that people have to be trustworthy, or they face the wrath of government otherwise.
Essentially, in short, the idea that humans have regularly been trusting people based on asking their close compatriots who they can vouch for doesn't seem to accurately model how trust develops in societies for most of human history--you either have vouching based on direct observation (in small societies), vouching based on identity (larger non-state societies), or centralized authority vouching for people (state societies, including today).
Humans are generally considered to have been around for only ca. 300'000 years.
Also, the interpersonal web of trust is IMHO quite clearly failing in the age of mass media, and not just recently.
Sure, but what do we do from here? Attempt to augment it with technology so that it starts scaling again, or give up and replace it with something centralized?
> We already have a functioning system for delegating domain ownership in a hierarchical fashion
Well, yeah. DANE should have won. NameConstraints would be nice tooo, but who's going to make CAs use it?
And FFS, clients could at least check CRLs once in a while.
DANE replaces this system with one that has the same basic cryptographic properties, but in which you can't eject bad actors --- or even monitor them with a mandatory transparency log. It's strictly worse than this.
WebPKI can't eject the bad actors that would cause trouble in DANE either. If your TLD conspires against you, they can certainly establish domain control and get a properly issued WebPKI cert.
Visibility through Certificate Transparency logs would at least make it a matter of record that it occurred, though.
The WebPKI semi-routinely kills certificate authorities. The point is that you can't kill a TLD operator.
> The point is that you can't kill a TLD operator.
I agree, you can't kill a TLD operator. But WebPKI relies on TLD operators to establish identity, so I don't see what we gain from WebPKI, other than mandatory CT logs, and not having to faff about with DNSSEC :P ... both of those are pretty valuable, of course.
I'll put it another way: why do we want a CA to tell us that someone controls a domain, when the TLD could tell us? The TLD is authoritative with regard to who controls a domain; CAs can only report on what they have observed.
Today's incident isn't a domain binding issue; it's a key mishandling issue. Elevating the role of TLD operators here just increases the number of organizations we have to trust to handle keys properly; worse, as we agree, those new keyholders can't have their authority revoked.
In the main, attacks on the DV process aren't protocol attacks; they're simple ATO attacks on registrars. The DV part of the WebPKI is certainly gross and could be improved, but evidence from the last 20 years says pretty clearly that the DV process itself is not the most urgent risk in the system.
My first hope is that we continue to ratchet security requirements for CAs upwards, and in doing so wash more and more incompetent operators out.
My objection is your choosing to build mechanisms that force us to make sure your "we" can "continue to ratchet security" instead of building mechanics that distribute trust and enforce it technology.
I don't care if you write 100,000 or 1,000,000 words: there's nothing wrong with DNSSEC or security that rides on it in principle except that it distributes trust decisions that have been centralized in those objecting to DNSSEC.
It is very bad for the internet to depend for its security on a tiny number of individuals to make global judgements about who's a bad actor.
Your counterproposal is simply to expand the number of individuals who get to make those global judgements!
How so? In the current regime, I can obtain a certificate automatically via ACME. if I control the domain, so to retain control over myexample.com, I need to trust the DNS infrastructure up to my TLD and trust every CA, including the "my secret key is return 4" crowd
In a DANE world, I need only trust the DNS authorities, and there are ways to boost trust in, e.g. the dot-com key management. (Could even go full ring signatures if we wanted.)
Someone over on foo.wtf.lol can't hurt me.
Two problems with this: first, it describes a spherical cow universe where there is only DANE; on Earth 1, widespread deployment of DANE would add to the number of trusted authorities, not restrict it. More importantly: it assumes a universe in which everyone leaves .COM, which is controlled by the US IC, and somehow finds a TLD that can be trusted both as a domain operator and a CA. When that institution is inevitably suborned by the IC, there will be no recourse, except to migrate like a hermit crab to some new TLD and wait for it to happen all over again.
Meanwhile: this CA is likely to be killed outright for this fuckup.
What stops our adding a record that requires DANE and a conventional CA? (Preload it like HSTS, if you'd like, to prevent US IC downgrade attacks if people stupidly configure DNSSEC to fail open.)
This way, fuckups like this CA can't hurt me because they couldn't break DANE and IC couldn't hurt me (more than they can already) because WebPKI would have to approve too.
I don't think the CA death penalty addresses the problem. Just as in the physical world, the death penalty isn't a perfect deterrent.
The fact that DANE doesn't work on all network paths and it's hard to predict which ones will fail. You have a preference not shared by ordinary computer users, that broken network paths should be loudly broken and not best-effort.
If you care this much about your root set, consider having Claude roll you a Chromium or Firefox with a user-controllable set of roots that you ruthlessly prune and manage. Unlike DANE, which simply isn't going to happen for a whole variety of reasons, your personal root program (I gift that term to you if you want the coinage) actually could help people with your preferences.
> It's strictly worse than this.
Maybe if people implemented name constraints and the CA/B forced CAs to use them. But not as it is; unconstrained sub-CAs are a scourge on the Web PKI.
Name constraints are intended for organizations which want to issue publicly-trusted certificates for their own domains. Name constraints are already widely supported and name constrained sub-CAs exist for this purpose in the WebPKI.
eMudhra is not such a case. They are a root CA that issues certificates to the general public all over the world. In fact, they are now the preferred vendor for numerous US universities via InCommon[1]. There is no way to meaningfully name constrain them.
[1] https://incommon.org/certificates
Can you share any insight about how an Indian CA became the preferred vendor for US universities despite the existence of LetsEncrypt? I know this is kind of a dumb question.
I had a friend who worked in university IT and I got the impression that there was a strong desire to use "supported" products (in general, not just certificates) even if it cost money and was worse than free alternatives.
Interestingly, InCommon used Sectigo prior to this year. I'm guessing eMudhra costs less.
I like constraints too, but listen carefully to EKR and 'agwa on this, because a lot of things that seem like straightforward fixes are much more complicated in reality. The most important thing I want from the WebPKI is expensive entry and the death penalty for fuckups, and I think we basically have that now.
It has the property of aligning cryptographic authentication with the domain name hierarchy. Yes, you trust the hierarchy. You trust LE too, so what? We need to move away from any authority being able to authenticate anyone, and DNS naturally aligns with the goal.
You've been vocal for a long time against DNSSEC deployment too for no discernable reason.
Why are you against every single thing that tries to make DNS useful for securing anything?
Do the words "no discernible reason" just mean something very different to you than they do to me? Because I've written somewhere in the vicinity of 100,000 words about the problems of DNSSEC over the last 20 years. My written DNSSEC critique predates HN itself.
This particular issue has nothing to do with "aligning authentication with the name hierarchy". The exact same thing that happened to EMudhra could happen to any TLD operator. The difference is that we have essentially no recourse when that happens, and far less ability to detect malfeasance.
People have debated DANE endlessly here, so I'm not sure how much more I have to add at this point. However, browsers, at least, do swiftly handle revocation for CAs, as indicated in this comment: https://news.ycombinator.com/item?id=50167244. They also handle EE revocation to varying extents, with Firefox having complete coverage and other browsers having partial coverage.