NSA tries to weaken mlkem standardisation?
nsa.2026.action.cr.yp.to
nsa.2026.action.cr.yp.to
This tactic is explicitly called out in RFC 7282, and named as a "degenerate", "pathological", and "dysfunctional" state for the working group to be in. Shame on DJB for attempting to drive the working group into terminal dysfunction.
I'm out of fresh tin-foil hats as well, but it would not surprise me in the least if any government was actively engaged in weakening security and privacy protections.
Literally look at what they are all doing in almost every sphere. The current political zeitgeist is all about automated surveillance everywhere. The motivations are worn on their sleeves.
https://en.wikipedia.org/wiki/NOBUS
DES key-size weakening is consistent with NOBUS (given the computational dominance of the US at the time). DUAL_EC_DRBG is consistent with NOBUS. DES S-box strengthening (vs linear/differential cryptanalysis, I forget which) is also consistent with NOBUS.
There have been *no* proposed mechanism that would allow NSA to have a NOBUS-style attack against ML-KEM.
Separately, this RFC (pure ML-KEM) is marked "recommended to implement = N". It is highly likely all browsers etc will use hybrids. In certain areas (say hardware) it is not free to use a hybrid. You all of a sudden need both a SHA2 and SHA3 implementation, for example. Some organizations that view the threat of quantum computers as more credible may also not want to drag around the ECC component (which is known to be broken, once a CRQC appears. Google and the US government have publicly stated concerns this may occur within the next ~5 years after recent QC breakthroughs).
Simply reducing the complexity of the standard to "pure ML-KEM" could already be considered enough "NOBUS" to be workable, such that the focus of attacks can be only on it (and bonus NOBUS if weaknesses are already known).
Sure, it's not completely free, but the hardware and implementation points seem relatively minor. Once CRQC exists the capacity will certainly not be unlimited, so there will surely still be use for encryption using ECC and "dragging it around" is not so bad.
As for the NSA, yes, he documents what is happening pretty well, though it's done through above board channels rather than envelopes full of cash. He quotes an NSA person as saying hybrid protocols are unlikely to receive approval for use in government systems. That is, "if you want the government to buy your stuff ($$$$$), it better not implement hybrid". Similar to Dual_EC_DRBG.
Anyone who attributes public motives or activity or blame to "the NSA" automatically gets dropped into the "conspiracy theorist" bin, as far as I'm concerned.
The Dale Gribbles of the world are not a particularly common character to meet in real life but that's the image evoked by "conspiracy theorist" no matter who the pejorative is aimed at, no matter the arguments they make nor the evidence they present. "Conspiracy theorist" is a thought-stopping, ad hominem cliché, with no place in serious discussion, yet it is widely used in exactly that; the possibility of conspiracy itself is what is eschewed in such discussions.
I would love to see any sources on this claim.
A lot of "conspiracy theories" end up being true in some vague way; "the NSA is spying on all of us", yeah, that was true. The NSA is using satellites to read our thoughts? Not so much. Still, people will point to things like the Snowdon leaks to prove that the US government cannot be trusted (which is true) and therefore all the other claims that people make are also true.
The reality is that most conspiracy theories are impossible, either from a technical sense or a logistical one. The extreme examples, like "the earth is flat and the governments are hiding it" or "COVID isn't real and the vaccine is going to kill everyone but every government and doctor on earth is secretly in on it" get shrugged off as "well, not THOSE ones obviously", but most of the rest I've ever seen are also completely unbelievable.
Here's the thing: anyone can come up with a theory and stitch together the most circumstantial "evidence" to "prove" it, combining misinformation, misunderstanding, and misrepresentation to produce something that feels like it could be true on its face if people don't do any real digging, and most don't. I've yet to see a "conspiracy theory" backed by any actual hard evidence; they seem to entirely spring from an overactive imagination and are then "justified" and "proven" by finding other facts to fit the narrative retroactively.
Is there clandestine activity? Absolutely. Are there groups of people trying to manipulate situations and lie to the public for their own gain? Almost certainly. Do people with unsourced, unproven conspiracy theories make it easier for governments to get away with whatever they want because the rampant proliferation of crackpot theories allows for a convenient smokescreen whenever the truth starts to come out? Also yes.
Even if only 10% of conspiracy theories are true, which they are not, the people repeating them do more harm than good by doing so in a way that discredits themselves and others.
He maintained that it was the most secure option available, which was technically true until the technology around mail transfer started improving with things like SPF records. qmail didn't support them and wouldn't support them, patches weren't accepted, and the only way to use things like SPF (to reduce spam) was through unofficial community patches that could never be upstreamed.
qmail was far better than sendmail at the tiume, and honestly it probably still is to a large degree, but, like forcing users to change their password every week, it was a case of security being so tight that users had to break it in order to make the system functional.
All this to say that, while DJB is undoubtedly insightful and intelligent, I'm wary of any of his claims of 'this isn't secure enough' because of his past history of making things so secure as to be inflexibly unmanagable.
But call me surprised that Filippo Valsorda and Rich Salz supported the NSA here.
What's the steelman of djb's position and why does it fail scrutiny? To the uninitiated, it sounds like his preference for the hybrid classical/BQP scene is prudent given the marginal computational burden.
But I'm guessing, it would be better if an expert weighed in with details.
The kindest reading of DJB's position is simple: pure ML-KEM is strong against fewer potential future scenarios than hybrid algorithms are, so people should use hybrid algorithms instead.
And in fact, I agree with this statement! The marginal cost of hybrid algorithms is very small, and the extra safety provided by being hybrid is (in my opinion) slightly larger than that cost. My cost/benefit analysis says that hybrid is the way to go for most people.
However, there are two major flaws with DJB's actual argument:
1. He's making the jump from "most people shouldn't use pure ML-KEM by default" to "the ability to use pure ML-KEM shouldn't be standardized at all". He's making overblown assertions about the power and meaning of an Informational IETF document, and using those to attempt to prevent simple interoperability standardization. Just because I think hybrids are the safer default in general doesn't mean that pure ML-KEM should be verboten; people deserve options, and the role of this document is to provide interoperability instructions for that option.
2. He's resorting to character attacks to imply that ML-KEM simply isn't safe at all. He frequently points to support from NSA and GHCQ as evidence that ML-KEM has been suborned in the same way as DUAL_EC_DRBG, despite widespread agreement in the cryptography community that the ML-KEM parameter space simply doesn't allow such attacks. He frequently points to support from cryptographers at Google and Cisco in the same breath, implying that they too are in the pay of the NSA. He's refusing to acknowledge that his implications that ML-KEM is unsafe imply that he should also oppose standardization of ML-KEM hybrids.
So while there's a kernel of truth to DJB's argument, I strongly believe that he has both taken the conclusion too far, and taken his argumentation tactics too far. It has lowered my respect for him as a person even further than it already was by his defense of Jacob Applebaum.
DJB saying so sounds good enough for me, considering the alternative. Dual_EC_DRBG anyone?
Ironically, it's the incredibly weak pro-standardization arguments that appear to be the most convincing evidence that the proposed standard is actually an instance of NSA meddling.
That post says very clearly at the beginning that hybrids are the preferred approach right now.
No one except the NSA actually wants a non-hybrid.
Which raises the question what is the NSA up to.
Especially since the NSA has a mission statement, a track record, and a billion dollar budget to subvert other peoples cryptography. When they aren't beyond transparent why should anyone give them the benefit of the doubt?
- European group could not be infiltrated by a state-actor with 100billion/y budget and a history of doing so?
- NOBUS today would not be secret in the algorithm but a quantum algorithm/device. Just a month ago HN was getting flooded with "PQC is probably required by 2030".
Ironically, this (delaying PQC rollout/standardization) is arguably what DJB has been doing the ~decade, and what his current post is doing.
I was under the impression certain dedicated single-algorithm quantum computers might be much easier to build; allowing you to attack some construct but not yet do full Shor.
PS I'm not saying that's whats happening. Just trying to nail down the scope of what is possible (not plausible).
Even dedicated single-algorithm quantum computers aren't magic. Given a dedicated single-algorithm quantum computer for attacking ML-KEM, the best current cost estimate we have for it is undoubtedly slower than the classical attack. Attacking ML-KEM quantumly is thought to take exponential (quantum) time. this is (clearly) not the case for ECC.
Could you elaborate?
Again explicitly, this is not the main RFC for PQ TLS, which details a hybrid construction. This is an RFC with "recommended to implement = N" marked about how to do PQ TLS 1.3 in environemnts where hybrids are too expensive, for example hardware where it necessitates both a SHA2 and SHA3 impl.
I think the argument boils down to this, yeah.
I am not a cryptographer, nor I’m participating in IETF (yet :), but he does make a good argument on why sticking with a hybrid for the time being makes sense (in between of all the NSA tinfoil hat stuff). And from an outsider point of view, publishing this as an RFC would somewhat legitimize using ML-KEM alone even though it’s marked as Recommended: N. (I would rather prefer waiting until we can publish it as Recommended: Y instead!)
If there are environments where ECDHE-MLKEM is really that much more expensive than ML-KEM alone, could we figure out another hybrid construction instead? E.g. one that only uses SHA3, if that’s the problem.
Saying that there's no "Nobody but us backdoor" to prove there's *no* backdoor of *any kind* is clearly naive at best, dishonest at worst.
As an example - if there's a weakness that affects 50% of keys (replace with whatever hypothetical number), NSA can make sure it doesn't use those affected keys but still retain the ability to decrypt 50% of everyone else's communications. And using the entropy analysis from this post, that would require 1 bit hidden in the parameters which is clearly within the entropy budget.
Dual_EC_DRBG is the shape of backdoor that would need to apply here: even if you knew the structure of it, you would need an additional private number to attack it. Recall the Juniper vulnerability where another threat actor simply replaced the EC public key used by Juniper's Dual_EC implementation.
NOBUS without some mathematical assurance that, even should an adversary discover the same break through, they cannot decrypt the same traffic would be too risky when you consider the NSA's self interest and dual mission.
That’s not a NOBUS backdoor. It’s a different type of backdoor and I’m pointing out that proving there is no NOBUS backdoor doesn’t mean there’s no other backdoor.
It’s a counterexample that I came up with in 5 minutes, not a proof that it’s useful to a state actor.
And why would they still be migrating top secret communications towards the algorithm they (NOBUS or not) have a backdoor in?
That doesn't sound very COMINT to me.
The context of this is that D.J Bernstein has been moderated 7 times from the mailing list for repeated unprofessional and disruptive behavior: https://mailarchive.ietf.org/arch/msg/tls/lON9lKptnJ6ccq2-I1...
He's also responded directly to criticisms: https://nsa.2026.action.cr.yp.to/guide.html
By comparison the so-called "unprofessional behavior" cited is the sort of subjective abject procedural bureaucratic bullpucky often used to shut down inconveniently correct criticism.
There are already codepoints assigned for MLKEM 512/768/1024 (0x0200, 0x0201, 0x0202) and nearly every major library supports it already:
- OpenSSL (ML-KEM-512/768/1024)
- BoringSSL (ML-KEM-1024)
- NSS (ML-KEM-1024)
- AWS-LC (ML-KEM-512/768/1024)
- Rustls (ML-KEM-768/1024)
- s2n-tls (ML-KEM-1024)
- Bouncy Castle (ML-KEM-512/768/1024)
- Botan (ML-KEM-512/768/1024)
- GnuTLS (ML-KEM-768/1024)
- WolfSSL (ML-KEM-512/768/1024)DJB supports the use of ML-KEM in TLS, but he correctly says that using only ML-KEM is unwise, because absolutely nobody can guarantee that no method to break ML-KEM will be discovered in the next years, as it already happened with the algorithm that was preferred before ML-KEM, until it was broken a few years ago.
All of these libraries also support the hybrid forms, which have different codepoints and are used by default. Nothing in the IETF process has any bearing on this.
The Recommended flag is set for X25519MLKEM768. It is not set for any of the pure PQ key exchanges.
https://www.iana.org/assignments/tls-parameters/tls-paramete...
Then don't use it if you don't want to? Use hybrid? ML-KEM (CRYSTALS-Kyber) is a NIST standard, FIPS 203:
* https://csrc.nist.gov/pubs/fips/203/final
* https://en.wikipedia.org/wiki/ML-KEM
There's also HQC, FIPS 207:
* https://csrc.nist.gov/presentations/2025/fips-207-hqc-kem
Unless you're doing government work, there's no reason why you'd be force to use Officially Approved™ algorithms:
* https://en.wikipedia.org/wiki/Commercial_National_Security_A...
If the (US) government wants to use (allegedly) compromised algorithms with-in itself that's up to them. The rest of us can use whatever we want.
https://mailarchive.ietf.org/arch/msg/tls/SXo4iVmp0ng_vi57ce...
https://keymaterial.net/2025/11/27/ml-kem-mythbusting/
Additionally, I wrote my own blog posts recently that toucbed on the subject.
Signatures: https://soatok.blog/2026/04/13/hybrid-constructions-the-post...
Threat modeling but also KEMs: https://soatok.blog/2026/06/30/soatoks-informal-guide-to-thr...
The main industry that's hamstrung by an RFC being blocked are telecom companies (e.g., Verizon) who by policy need an RFC and also have other regulations.
I prefer hybrid KEMs, but support publication because getting those companies onto PQ is harm reduction against Harvest Now, Decrypt Later (HNDL) attacks, and ECDH doesn't help if our confidence in the security of ML-KEM turned out to be wrong.
Demonstrating a 'pattern or practice' is the legal standard one has to meet to bust qualified immunity and shift the burden of doubt on to authorities, so I'd say it goes a fair amount past 'reasonable suspicion', which in a court is itself enough to issue search warrants.
Additionally, the main authors behind ML-KEM are all european. The design of ML-KEM is "very boring", in the sense that it's essentially the scheme that most (lattice) cryptographers would have suggested. There were 2 other NIST PQC schemes that went very far (New Hope and Saber) that were essentially the same scheme (there were minor technical differences, but it's really not that big).
> use pure ML-KEM > hybrid ML-KEM.
the current document is instead to say
> If you are in a setting where you REALLY want to use pure ML-KEM (though we explicitly recommend you do not do it), this is the standard you would implement against.
It is also technically inaccurate. The whole argument hinges on it being negligible cost in all environments to do hybrids. This is explicitly false. See this message on the TLS-WG on explicitly this point
https://mailarchive.ietf.org/arch/msg/tls/_9i3uIVDQ3pDRswpm9...
A large list of pros/cons detailing a question that isn't being debated that is technically inaccurate is what I would expect from an LLM, not from a competent cryptographer. I am unsurprised to see it from DJB given his behavior in the last decade regarding PQC.
>> If you are in a setting where you REALLY want to use pure ML-KEM (though we explicitly recommend you do not do it), this is the standard you would implement against.
Ah! This corroborates a point I made in another comment:
> Especially given the "not recommended to implement" part of it; something (CNSA2) tells me that this "not recommended" will be widely disregarded in favor of "but it's a standard" to the point that an explicitly-known-to-be-weaker implementation becomes one of the, if not the, most deployed implementation in practice. Which is also the NSA's MO.
https://news.ycombinator.com/item?id=48773930
The "setting where you REALLY want to use pure ML-KEM" is when you are are a US government contractor and therefore required by your contract to follow the NSA's recommendations (CNSA2). Given the discussion, it seems a bit dishonest not to mention that part, no?
I cannot see how any true "expert" would have the courage to claim that in a cryptography standard it is admissible to accept risks that cannot be quantified.
For the variant supported by DJB there are no risks, while for the variant supported by NSA nobody can estimate the risks.
It is as simple as this, so it is weird that there are people arguing about a decision that should have been non-controversial.
The variant DJB suggests there are explicit risks. For example
1. both ECC and ML-KEM can be broken (obviously)
2. additional code complexity could increase the LoC of teh crypto implementation, making it more plausible there are implementation bugs
regardless, this is a red herring. Nearly all cryptographers still support hybrids!!! The current RFC is *not* about "use pure ML-KEM". It is instead about "if you're going to use pure ML-kem (and we explicitly recommend not doing so), here is how to do it in a standardized way".
The people arguing about this decision don't even know what the decision being made is in the first place.
This makes the push for the standard far more suspicious. Why is it so important for this to be a standard if it is explicitly not recommended to implement at the time of standardization? The typical benefits of a standard would be to avoid disparate implementations, which seems fine for something that isn't recommended to implement.
On the other hand, lots of folks from the NSA coming out (covertly, in this hypothetical context) in support of a weak standard with dubious arguments is... the NSA's modus operandi. Additionally, the fact that the standard is being proposed so US government contractors can checks notes meet the NSA's recommendations(!!) is another reason to suspect the NSA's involvement (it seems weird I even have to write that). Especially given the "not recommended to implement" part of it; something (CNSA2) tells me that this "not recommended" will be widely disregarded in favor of "but it's a standard" to the point that an explicitly-known-to-be-weaker implementation becomes one of the, if not the, most deployed implementation in practice. Which is also the NSA's MO.
Edit:
Now that I think about it, recommendations like CNSA2 also support the NSA's spying capabilities. A single standard (or small set of standards) is easier to crack and exploit than many bespoke implementations. Granted, that's a bit of a weak argument since many bespoke implementations are likely to have their own vulnerabilities. The reason the NSA might still prefer standards be used is that a bespoke implementation will more likely have a bespoke exploit, meaning they can't use already-developed exploits and will have to spend time making one.
This isn't a standard.
It's not on the standards track! Words mean things!
But to answer your question: some industries need an RFC. FIPS 203 also doesn't specify how to use ML-KEM in TLS.
Think phone companies.
The TFA has nothing to do with ML-KEM, but only about how to transition from the current algorithms to post-quantum algorithms.
For now, it is completely unknown how secure ML-KEM really is, because it is too new. For many complex cryptographic algorithms a decade or even a few decades have been required until someone discovered how to break them. The predecessor of ML-KEM, SIKE, has already been broken. Perhaps nobody will break ML-KEM, or perhaps it will be broken in a couple of years.
The only risk-free strategy is to use both ML-KEM and the current key exchange algorithm. This adds a negligible cost, because ML-KEM is much more expensive.
Therefore I agree with DJB about this, because I never bet that the worst case will not happen. Any good design must work fine even when the worst happens.
ML-KEM is not new. It's hardness is based on MLWE. LWE (a slightly harder problem) has been around for 20 years. Attacks against it stablized to be 2^{cn} time maybe 15 years ago. The value of c has been stable for nearly 10 years.
MLWE is mildly different, but still from > 1 decade ago. The only improved attacks are ~8x faster (and they are relatively naive). It has been a very popular cryptanalytic target since it was suggested (LWE has been dominating cryptography for the last >10 years).
It does not add negligible cost in all settings. In hardware, a hybrid scheme requires an implementation of SHA2 and SHA3. This is expensive.
Any good design must not do fine when even the worst happens. When we did the AES competition, we did not combine AES with DES (or 3DES) in case AES was weak.
This is beside the point though. Most cryptographers would still recommend the hybrid over pure ML-KEM. This RFC (for pure MLKEM) is marked "recommended to implement = N". It is purely for settings where the implementors independently want to use pure ML-KEM for some reason. In these settings, they should implement it against some standard, so it is interoperable.
That's exactly how it was with Dual_EC_DRBG.
E.g. https://www.schneier.com/essays/archives/2007/11/did_nsa_put...
I don’t understand why the NSA was so insistent about including Dual_EC_DRBG in the standard. It makes no sense as a trap door: It’s public, and rather obvious. It makes no sense from an engineering perspective: It’s too slow for anyone to willingly use it. And it makes no sense from a backwards-compatibility perspective: Swapping one random-number generator for another is easy.
My recommendation, if you’re in need of a random-number generator, is not to use Dual_EC_DRBG under any circumstances.
Except as we know now, after the standardization NSA was able to persuade/bribe vendors to implement it.
RSA is still a viable cryptography vendor, after accepting money to backdoor their product for paying customers. The standardization gave them a fig leaf of plausible deniability. Honest mistake, could happen to anyone, right? If they had needed to implement a "non-standard" backdoor, or if it had been officially struck from the standard, it would have been a lot harder to row away from.
For DUAL_EC_DRBG, the mechanism that could yield a backdoor was known pre-standardization. To get the backdoor RSA had to specifically use government chosen parameters.
These are not new concerns. If even a candidate backdoor had appeared in ML-KEM (similarly to how DUAL_EC_DRBG was), it would be a very different story. But nobody has ever even suggested something might be off!
So no, it's not exactly the same as DUAL_EC_DRBG. Different things are in fact different. Note that there are similarities to DUAL_EC_DRBG in contemporary cryptography. Russia has a block cipher Kuznyechik that has some very fishy structure in its S-box. We don't know how such structure is exploitable, but I would bet money that it is. Despite not being able to see an attack, we can see that things seem off in a concrete way. Nobody has *ever* suggested that for ML-KEM.
Wanting to standardize it's use without the secondary layer of protection provided by existing algorithms over the objections of a well known cryptographer counts as a hint to me.
In the same way that paying RSA to make Dual-EC DRBG the default RNG in it's security products when it was newer and more expensive than alternatives was a hint.
1. the payment to RSA (in 2004) was secret. So it could not have been a public indication of a problem, as it was not discovered until nearly a decade after it happened (in 2013, when it became public)
2. the problematic part of DUAL_EC_DRBG (the "hint of a backdoor") I was mentioning was known pre-2004.
blind paranoia is not a rational approach to cryptography. I say this as someone who prefers hybrid schemes! I just don't think it is sensible to attempt to "ban" the usage of pure ML-KEM by not standardizing it. It won't work! It'll just increase the risk of non-interoperable implementations.
2: And? You're missing a second part to this statement. Did you intend it to support some conclusion?
> blind paranoia
Characterizing criticism this way, instead of listening, internalizing, and adjusting your position is exactly why DJB's references to previous NSA interference stick. Y'all don't just have technical differences, you're going for character assassination. DJB has been consistent and explicit about the technical nature of his objections. I find his prose on the matter clear and well reasoned.
Your arguments seem disjointed, unorganized, specious, and lacking, in comparison, and less credible for the way you respond.
> I just don't think it is sensible to attempt to "ban" the usage of pure ML-KEM by not standardizing it. It won't work! It'll just increase the risk of non-interoperable implementations.
I think it's entirely reasonable to dissuade people from building non-hybrid systems during a transition period, and refusing to standardize them is an entirely reasonable way to signal that people shouldn't build or trust such systems during such a time, even stronger than a recommended_to_implement = N. No one has attempted to "ban" anything, so that's another gross mischaracterization.
So all the companies who want to sell anything using TLS to the government want to standardize this, so they can be CNSA2 compliant.
Everyone already supports this in major libraries; but some folks feel they need an IETF RFC specifying it.
(I don't have to comply with CNSA2 so I might have details slightly off)
Because this would seem pretty stupid, i.e. to disqualify something that supports both post-quantum algorithms and previous algorithms.
I have looked just now at CNSA2 and it only says that post-quantum algorithms should be used exclusively for key exchange and digital signatures after 2033.
So during this 7-year transition period it should be normal to use both post-quantum and classic algorithms, even based on what CNSA2 says.
Moreover, even "exclusively" can be interpreted in various ways, i.e. it can also be interpreted that there should be no key exchanges/digital signatures that do not use post-quantum algorithms, but without forbidding them to also use other algorithms, because such a prohibition does not make sense.
https://blog.cr.yp.to/20220805-nsa.html
> Some people seem to be unable to rationally consider the possibility that NSA is sabotaging post-quantum cryptography. I've heard people saying, for example, that submissions to the NIST Post-Quantum Cryptography Standardization Project (NISTPQC) were publicly designed and evaluated by top experts, and that NSA can't have bribed the submission teams. > > Let's look at the facts.
Note that the authors of ML-KEM are overwhelming European.
He just pointed that the predecessor of ML-KEM (SIKE) has already been broken. Because ML-KEM is also very new, there is a non-negligible probability that it will also be broken in a few years.
It is very simple to guard against this, by using both ML-KEM and the currently used elliptic-curve Diffie-Hellman algorithm.
ML-KEM is much more expensive than the current algorithm, so using both does not increase much the cost.
I do not see any flaw in his arguments, while anyone who says that ML-KEM should be used alone is making a bet for which there exists no justification, i.e. the risk is extremely high and the reward is extremely low.
In cryptography bets must be done only when the odds are extremely favorable, which is not the case for the proposal criticized by DJB.
https://mailarchive.ietf.org/arch/msg/tls/SXo4iVmp0ng_vi57ce...
Also, https://keymaterial.net/2025/11/27/ml-kem-mythbusting/
ML-KEM is not "very new" compared to the age of other algorithms historically deployed.
The hardness assumption from ML-KEM is from 2005 (in teh algebraically unstructured case. The biggest speedup known due to algebraic structure is ~3 bits, e.g. 8x speed improvement). It has taken exponential time to attack since then. Instantiating a standard ~20 years after introduction is slower than what we did with RSA, or with elliptic curve cryptography.
Therea re settings where hybrids are not free, for example hardware. The standard hybrid suggestion (XWING) would require hardawre to implement both SHA2 and SHA3. See this recent TLS WG post detailing this
https://mailarchive.ietf.org/arch/msg/tls/_9i3uIVDQ3pDRswpm9...
And who is making that 'ML-KEM alone' argument? AIUI, the IETF currently recommends hybrid.
It's on the US government talking to itself that would maybe do non-hybrid:
* https://en.wikipedia.org/wiki/Commercial_National_Security_A...
If the US government wants to compromise itself… then fine?
Also, HQC (FIPS 207) has been chosen as a backup / alternative to ML-KEM (formerly "Kyber"):
* https://www.nist.gov/news-events/news/2025/03/nist-selects-h...
https://en.wikipedia.org/wiki/Kuznyechik#Cryptanalysis
(the work by Perrin that is mentioned is what I'm referring to).
The (pure) mlkem standard is also marked "recommended to implement = No". people are interested in implementing it. The IETF can't change that. They can try to ensure such implementations are interoperable though.
> Recommended: N
* https://datatracker.ietf.org/doc/html/draft-ietf-tls-mlkem-0...
Pure, non-hybrid implementations of ML-KEM (FIPS 203, previously "Kyber") are already in the major crypto libraries. So given code is being written, there can either be an IETF document available to ensure interoperability or not: the IETF would prefer interop it seems.
The list of IETF-recommended TLS algorithms can be found under the (sortable) "Recommended" column of officially registered TLS parameters:
* https://www.iana.org/assignments/tls-parameters/tls-paramete...
The list is currently: X25519MLKEM768, x448, x25519, secp384r1, secp256r1. The document being discussed/debated would not change it.
Maybe giving this thread more visibility here than it wants but ...
https://bsky.app/profile/filippo.abyssdomain.expert/post/3mp...
(Personally it seems so so unacceptable to me to accuse so many good hardworking people of such bitter conspiracy.)
The linked piece is not representative of the broader cryptography community. ML-KEM is fine.
Someone elsewhere in the thread mentioned downgrade attacks. I presume if you wanted, on either the client or the server, you could disallow pure ML-KEM if you didn't trust it, preventing this vector.
I don't know much about the hardware space - what do you make of the author's post that there hasn't been an articulated need for pure PQ encryption, where the device couldn't afford ECC.
https://mailarchive.ietf.org/arch/msg/tls/_9i3uIVDQ3pDRswpm9...
In general most cryptographers don't do hardware. Most cryptographers would do something more conservative. I personally think what chrome is doing with the XWING combiner is fine/sensible.
I do NOT think that stirring up so much trouble over pure ML-KEM is fine/sensible. Especially since it comes after nearly a decade of trying to disrupt the post-quantum transition by DJB (his behavior around the NIST PQC competition). It has made me view him as a bad actor in this space, which is a shame given his previous positive contributions to the field.
RE downgrade attacks: you're right. For some prior downgrade attacks (TLS 1.2), see e.g. LogJam
https://weakdh.org/imperfect-forward-secrecy-ccs15.pdf
My (admittedly hazy) memory of the attack was that they were able to take a client + server who both support "export grade" crypto (say 512-bit finite field DH), then
1. force them to choose this (over cipher suites they may prefer more), then
2. solve the resulting instance before the connection times out, and then
3. use this to rewrite the transcript history to be consistent with one of the sides only supporting 512-bit DH.
This required both sides to support the 512-bit DH though, as well as having the ability to break it on the order of minutes. There is no public estimate that breaking ML-KEM is remotely that small. There are some public repositories of attack records on LWE, e.g.
https://www.latticechallenge.org/lwe_challenge/challenge.php
As you can see, the records are < dimension 100 (though that isn't the only relevant parameter). There might be some records elsewhere that push this mildly higher, but my understsanding is the record attack is
1. definitely << 200, and
2. attacks against ML-KEM have exponential time complexity, so scaling to ~500 would require quadratically more time. This is a huge difference in cryptography, e.g. the difference between a completely broken scheme (say complexity ~2^50-2^60) and AES (~2^120).
The use of dual algorithms is without doubt the prudent decision for a transition period.
ML-KEM is still too new for anyone to be able to claim that no way to break it will be discovered in the next few years.
This is supported by the fact that one of the algorithms previously proposed for standardization has already been broken, which was a surprise.
Because ML-KEM is significantly more expensive than the current algorithm, using both does not increase much the cost.
The arguments of DJB are perfectly valid, which is why at the previous meeting most people have voted like him.
I know very well everything that DJB has published during the last 30 years, many of which have been important advances in cryptography. Some of his work has been very influential in the development of "post-quantum" cryptography and he was one of the main promoters of the idea that such cryptographic algorithms must be standardized ASAP.
Moreover, I have also run continuously on my servers, 24/7, for about a quarter of century, various programs written by DJB, which unlike the majority of the programs that I have ever seen, have done very well whatever they were intended to do, without ever needing any updates for security problems or other bugs. Very few programmers, even among the best, can present such a resume.
I do not believe that "crank" is the right word to describe DJB. It is true that he has distinguished himself by an unwillingness to accept compromises, even when he was for various reasons in opposition with the US government, but I do not think that this is crazy. On the contrary, I believe that the world is how it is right now precisely because most people go with the flow and they are eventually willing to accept almost anything when opposing that appears to be too difficult. Things would have been much better if there had been more such "cranks".
The latest post to the list, as of this post, is supporting the anti-ecdhe side, with the reasoning being that there is no code written for ecdhe, which is obviously stretching the truth beyond reasonable doubt.
That’s pretty weak just stripping down the hybrid approach.
This is about a separate RFC with "recommended to implement = no".
If the IETF was trying to have these positions swapped, it would be consistent with DJBs post. It is not though. His post does not seem to be grounded in reality.
https://blog.cr.yp.to/20260221-structure.html
where he combats very well your argument.
For me, his argumentation seems far more grounded in reality than what you have said.
That blog post is written in a way that implies otherwise, namely that pure ML-KEM is being favored over hybrids for TLS 1.3. This is explicitly false.
Moreover many parts are technically false. In particular, the claim that hybrids are negligible cost in all circumstances is false in low-spec hardware, as it necessitates both a SHA2 and SHA3 implementation.
https://mailarchive.ietf.org/arch/msg/tls/_9i3uIVDQ3pDRswpm9...