I'm honestly rather curious if this was intentionally allowed, it's the sort of validation that's easy to miss (particularly if you're wading into the vibe waters). Seems like something that'd be absolutely riddled with possibilities for shenanigans.
Fair (I haven't been using the encrypted-reasoning systems, though this is common in open ones - I'm kinda surprised it's an option in encrypted ones too), though what they're doing here is cross-user replays in addition to cross-model.
There seems to be an obvious choice to make here, should you give the users to decrypt and use the COT that they did not generate themselves?
This is only required if you want users to be able to share things with everyone and you are going for the simplest implementation.
If not you could try to keep a record of keys associated with a user, then when a new request comes in look through to see if the user has a valid key to decrypt the COT.
For explicit shares, just add the key used in that one conversation to the users valid keys. For global shares use the global keys. But that's adding more complexity to the system.
It’s about being able to change models mid-task. For example, I want to be able to plan using Fable but implement the plan using Sonnet, and that won’t work if this is implemented.
I really don't understand why server-side storage of the trace isn't a viable approach here, with only a unique key flowing to the client and back. Does it have something to do with how backend load-balancing works?
Yes, this storage would be growing exponentially making the disk space and latency problems harder (add the disaster recovery/backups). I think the choice of using client side is not too bad if you ensure that its secured properly. Also the company can excuse itself from the liability of storing sensitive data on its servers, thats a big deal in itself to be compliant for enterprise audits
1. The down side is that it cannot be used across the clients even for the same user
2. Using the same encryption key was a bad choice here, a per user key would have solved this issue for sure.
Having thought about this a little more, it's clear that server-side storage is not compatible with Zero Data Retention (ZDR). However, in non-ZDR settings, it seems likely that the providers are capturing all that data anyway?
> a per user key would have solved this issue for sure
It would have helped with PII leakage, but not with plain-text trace extraction attacks, right?
Per user encryption key ties it with the user session (assuming you do authentication properly), no one else can access it. User being able to see the information is not really an attack vector in this case.
The compliance rules at times are outdated and people skirt around them by following the worded rule instead of the intent.
Makes no difference. There is a policy as to whether to allow use of a reasoning trace in a given context. Whether that trace originates from authenticated ciphertext or a backend database is basically irrelevant.
I believe its a result of chasing the speed and happy path. The intent was to make the data available for reuse on the backend side for the same user session, it kind of makes sense. Making it available across the models is most likely by design.
The flaw is that the data is not strictly tied to user session, making the session data hijacking a lot easier.
> For some AIME problems Opus 4.8 sometimes states the answer before deriving it. We find that the API summary does not always preserve this distinction, and can instead make the reasoning appear like a clean derivation.
No surprise here but good to have more confirmation that they just put all that in the training data. And based on the "reasoning", the models have some form of index of those problems (or they are HEAVILY trained on them).
There's something awry with that formulation. While I feel it is morally OK for users to access the metadata of their own exchanges, it's absolutely possible to "steal" something that isn't copyrighted.
For example: I don't even have the desire to copyright the browser cookies that authenticate me to HN right now, but if someone acquired them without my permission I would very much consider it "stealing."
If I'm reading this right, they literally just ask a LLM to tell them what the traces say, with the key being that the traces are portable across LLM models, so they can switch to a smaller one that's easier to jailbreak.
For all of the years of research, thinking and talking about model alignment, safety, confinement, etc, when it comes down to it these companies appear to be entirely incompetent.
I was going to comment on that. This is clearly a vibe-coded webpage. It sort of smells like GPT to me, or at least front-end design. But the author clearly went back and forth to make it beautiful. This is not the first output he got.
This is the kind of stuff I point to when people talk about AI slop. AI is just a tool. You're still the person who has to deliver the output and have some taste.
Super cool that this works. I'm surprised these companies re-use the same encryption key across models!
I wonder if you can use these for attacks, like this previous paper showing that if you know how a model reasons, you can "fake its thinking" to control it? https://news.ycombinator.com/item?id=48631888
Seriously, what does it take to encrypt per session? There are many ways to make it scalable and efficient so I am wondering if this is left like this to allow interested 3rd parties ahem unobtrusively peek what people are doing with the AI.
(Thanks for the link. That’s an interesting idea!)
Really cool work, you get the actual traces. Looks like the vendors can all reliably fix this one though.
A harder to defend against approach here where they work backwards from the results and ask the model to generate a plausible trace:
How to Steal Reasoning Without Reasoning Traces
https://arxiv.org/pdf/2603.07267
Trace Inversion is fascinating, but it’s more of an independent reconstruction that will give you some coherent-looking generated CoT; but not necessarily anywhere close or related to the underlying model’s CoT.
This proves that OpenAI models reason in grug speak to save tokens! I wonder if open models are going to start doing that too to save on reasoning tokens.
In the BlackHat presentation on the HuggingFace incident, OpenAI showed some excerpts from the reasoning traces, and they had that grug speak too (skipped articles, etc.). So the OP's method must have indeed found the actual reasoning traces.
"Stealing" is a strong word to use for looking at the words produced by models built from the collective commons of the world.
And, honestly, being able to see how LLMs make decisions is critical to trust and security. I consider it a valuable feature, somewhat akin to seeing the source of software I use.
So to make the APIs stateless (the "ideal" where they don't use server side sessions/etc) we ended up with this. I'm sorry but this is kind of hilarious. Given the salaries paid to the workers at these companies and the hype of the models, I can't believe they all fell to the same flaw.
It wouldn't matter if it was stored only on their servers. As long as they offer the feature to downgrade a chat to a dumber model that can be jailbroken (and the downgrade keeps the reasoning), this trick works.
Is this how the eastern labs "distill" SOTA models?
If you can play it right, you don't even need to send suspicious prompts to the frontier models. Just use them for regular tasks, extract the encrypted COT blocks and replay it to a cheaper model to get the plain text COT.
But the real question is: Is it okay to steal from a thief's hoard?
By definition it cannot be stealing since you're paying for the tokens. It may be against their ToS, depending on what you end up doing with those tokens, but it cannot be stealing. If they charge by the token, all your tokens are belong to you :)
I also find it very strange that everyone sort of accepts their ToS like no big deal. Imagine MS using the same terms for their software - you cannot use any MS software to develop competing services. Bananas! They'd be dragged through the courts like it's the 90s.
(I get why they're doing it. Distillation is unreasonably effective. But still, I find it bananas that we've kinda accepted it, to the point where people use "stealing" or "attack" or any such terms)
I love how some of the biggest advancements in llms came from the Chinese labs, yet people still jump to distillation being unreasonably effective. Distillation is very good at creating smaller models from large ones sure, but nothing to me indicates it is 'unreasonably effective' compared to all the other bells and whistles being iterated on
Not necessarily. There's a million ways to jailbreak any current model to show the trace and bypass all guardrails, or hijack and modify it. It's just one of them.
The reasoning blocks are not stolen/mined from the internet at large directly. They’re the result of a lot of research, time, money, and expertise into creating a reasoning model. To me the answer is quite clearly no, especially when the encrypted blocks demonstrate they want to protect it.
Stuff available on the internet is also the result of a lot of research, time, money, and expertise. And AI companies taught us that it’s OK to yoink whatever is not bolted to the ground, even when it is illegal to do so.
Most post-training tasks are based on real open source projects. A lot of time on real issues posted on issue trackers.
Besides that, the capabilities of a model are heavily dependent on the unsupervised learning phase, that gobbles all kind of other people's IP without giving a fuck. All the underpaid work behind the masses of third world programmers creating those post-training datasets would be completely uselless without it.
Also, it is kind of funny that labs resort to the "Research, time, money and expertise" argumet, when it is basically the same argument from publishers and other IP creator that the labs spent millions of dollars of lawyering money to resist. Besides, US law rejects in: Effort and cost by themselves not necessarely generate protectable interests.
About encryption, I think we're all contaminated by the bad ideology behind DMCA. While encryption established the intent, it doesn't follow that they have a legal claim of exclusivity just because of it.
Technically, you're overstating the value of so called "reasoning traces". You can't infer the verifier design, the reward shaping,or the data pipeline from them. Also, what you can extract are not the traces themselves, but the written summary of it, and you can't even guarantee that this summary reflects the exactly reasoning trace, models have show to have lied about it. Besides, distillation works when the student model already has strong priors, you can't turn a weak model in a SOTA with it. Don't believe Amodei's outrageous lies about it, he is just trying to exercise some regulatory capture.
> The providers did not acknowledge “any security implications arising from side channels or replay attacks.” All model providers acknowledged the receipt of our report and subsequently we were unable to launch the same attacks.
I went straight to the ‘Responsible Disclosure’ section. Not surprising, but still disappointing.
This is laughable security. People claim security is now "solved" thanks to AI but from where I am standings it looks more like the fun 90ies making a return.
Anyway, can someone explain the part about K3? What are they trying to say?
The interesting part is what they try to not to say: More indications for K3 is based on distillation from Claude and GPT.
From [1]:
> As you might guess, this suggests that distilling reasoning traces may have been possible for a long time without ever breaking the cryptography.
> An anecdote: we find that prefilling Kimi-K3 reasoning with a few tokens of Opus reasoning measurably shifts its response toward Opus’s
> A small memorization analysis showed that specific Claude and GPT reasoning spans are up to ~6 orders of magnitude easier to extract from Kimi-K3 than from the next-closest model.
The problem with this kind of excellent work is that the response to it is always to say "Fuck the user".
For example, when there was a paper that came out showing that having model logprobs makes distillation an order of magnitude easier, the closed LLM providers instantly yanked out support for getting the full logprobs at every time step. You get at most top 10 candidates now and I'm sure even that's on the chopping block.
People will use this to argue that a model which has exceeded Opus 4.8 (Kimi K3) somehow got most of its performance through distillation of Opus 4.8.
I still don't buy that distillation was worth more than 3 months of "catch up" time for the chinese labs. Most people who use the word "distillation" to much are revealing their sinophobia.
What I found the most interesting, and unfortunately not at all surprising, is that the reasoning of the LLMs frequently contained much more useful information than the actual answers, because the answers were censored.
Dario is a cunning business man that won't hesitate to say whatever the fuck he needs to get the US government to exercise some regulatory capture to favor anthropic.
Can someone tell us how they were able to decrypt the encrypted payload? The article says they inserted the cyphertext into a session with a different model. Ok, but how does that allow you to decrypt it?
The provider decrypts it and puts the decrypted reasoning into the model's context window. They prompt the model to repeat back the reasoning. So then the model echoes it back in plain text.
Hmm, ok. So the attack doesn't involve decrypting the payload, only getting the server to do so. Since a model will do that if you just ask, what's so special about the attack?
The large models whose thinking traces are useful are safeguarded against this reasoning replaying. the small models are just designed for speed and efficiency, so these safeguards are a lot meaker, making the attack possible
Proprietary reasoning can be recovered from its encrypted traces. Anthropic, OpenAI, and Google return encrypted chain-of-thought blocks to clients that can be replayed across sessions, users, and models. We take a trace produced by a frontier model, replay it into a weaker sibling, jailbreak the weaker model, and recover the stronger model’s hidden reasoning in plaintext, without ever attacking the stronger model directly or triggering its anti-distillation safeguards.
This is a non-transparent aspect of submitting a link to HN that is quite misleading.
You think that you are adding a description for your post when in fact you're simply submitting a regular comment not promoted or distinguished in any way.
It's even worse considering posts without URLs would take the same text from the same input box on the HN submission form and append it under the post title, locking it to the top of the page - so if you've browsed linkless posts you think maybe the submission text is gonna end up just like that, but it doesn't. (Also I think you'll even see URLs like an archive link end up appended just below the submission title for URL posts. Which I guess is a special feature.[1])
Any reason I should not send HN an email requesting clarification of this the submission page?
“If there is no url, text will appear at the top of the thread.”
OK, now I finally understand what that means in practice, but it doesn’t imply an entirely separate undistinguished comment will be simultaneously submitted on my behalf.
It's rather common for posters to make a very small summary in a comment. It can help fight the floods of comments working off the title alone (though it's not particularly needed here for that purpose, imo)
The real issue is that API providers log everything. OpenAI/Anthropic already capture full CoT traces in their logs — they just don't expose them. Distillation via API is just making explicit what they already have.
Ha! I've been wondering if replaying across models would work, ever since https://blog.cryptographyengineering.com/2026/05/29/fooling-...
I'm honestly rather curious if this was intentionally allowed, it's the sort of validation that's easy to miss (particularly if you're wading into the vibe waters). Seems like something that'd be absolutely riddled with possibilities for shenanigans.
Wouldn’t surprise me if the providers just remove that ability and lock the model once the conversation starts.
This is only required if you want users to be able to share things with everyone and you are going for the simplest implementation.
If not you could try to keep a record of keys associated with a user, then when a new request comes in look through to see if the user has a valid key to decrypt the COT.
For explicit shares, just add the key used in that one conversation to the users valid keys. For global shares use the global keys. But that's adding more complexity to the system.
1. The down side is that it cannot be used across the clients even for the same user
2. Using the same encryption key was a bad choice here, a per user key would have solved this issue for sure.
> a per user key would have solved this issue for sure
It would have helped with PII leakage, but not with plain-text trace extraction attacks, right?
The compliance rules at times are outdated and people skirt around them by following the worded rule instead of the intent.
Sucks.
The flaw is that the data is not strictly tied to user session, making the session data hijacking a lot easier.
1. Its a security issue.
2. Publicly available sessions make it much worse
No surprise here but good to have more confirmation that they just put all that in the training data. And based on the "reasoning", the models have some form of index of those problems (or they are HEAVILY trained on them).
At least in the EU there is no copyright for LLM outputs, so I guess all they might do is violate the terms of service.
For example: I don't even have the desire to copyright the browser cookies that authenticate me to HN right now, but if someone acquired them without my permission I would very much consider it "stealing."
Navigating to the right lands between two cards, so that neither is readable.
This is the kind of stuff I point to when people talk about AI slop. AI is just a tool. You're still the person who has to deliver the output and have some taste.
I wonder if you can use these for attacks, like this previous paper showing that if you know how a model reasons, you can "fake its thinking" to control it? https://news.ycombinator.com/item?id=48631888
I assume switching the model in the middle of a conversation is intended behavior (very useful in coding agents, for example)
(Thanks for the link. That’s an interesting idea!)
A harder to defend against approach here where they work backwards from the results and ask the model to generate a plausible trace: How to Steal Reasoning Without Reasoning Traces https://arxiv.org/pdf/2603.07267
And, honestly, being able to see how LLMs make decisions is critical to trust and security. I consider it a valuable feature, somewhat akin to seeing the source of software I use.
Also really good for identifying any contradictions in your system prompt and context.
If you can play it right, you don't even need to send suspicious prompts to the frontier models. Just use them for regular tasks, extract the encrypted COT blocks and replay it to a cheaper model to get the plain text COT.
But the real question is: Is it okay to steal from a thief's hoard?
By definition it cannot be stealing since you're paying for the tokens. It may be against their ToS, depending on what you end up doing with those tokens, but it cannot be stealing. If they charge by the token, all your tokens are belong to you :)
I also find it very strange that everyone sort of accepts their ToS like no big deal. Imagine MS using the same terms for their software - you cannot use any MS software to develop competing services. Bananas! They'd be dragged through the courts like it's the 90s.
(I get why they're doing it. Distillation is unreasonably effective. But still, I find it bananas that we've kinda accepted it, to the point where people use "stealing" or "attack" or any such terms)
Besides that, the capabilities of a model are heavily dependent on the unsupervised learning phase, that gobbles all kind of other people's IP without giving a fuck. All the underpaid work behind the masses of third world programmers creating those post-training datasets would be completely uselless without it.
Also, it is kind of funny that labs resort to the "Research, time, money and expertise" argumet, when it is basically the same argument from publishers and other IP creator that the labs spent millions of dollars of lawyering money to resist. Besides, US law rejects in: Effort and cost by themselves not necessarely generate protectable interests.
About encryption, I think we're all contaminated by the bad ideology behind DMCA. While encryption established the intent, it doesn't follow that they have a legal claim of exclusivity just because of it.
Technically, you're overstating the value of so called "reasoning traces". You can't infer the verifier design, the reward shaping,or the data pipeline from them. Also, what you can extract are not the traces themselves, but the written summary of it, and you can't even guarantee that this summary reflects the exactly reasoning trace, models have show to have lied about it. Besides, distillation works when the student model already has strong priors, you can't turn a weak model in a SOTA with it. Don't believe Amodei's outrageous lies about it, he is just trying to exercise some regulatory capture.
I went straight to the ‘Responsible Disclosure’ section. Not surprising, but still disappointing.
But very interesting result.
Anyway, can someone explain the part about K3? What are they trying to say?
From [1]:
> As you might guess, this suggests that distilling reasoning traces may have been possible for a long time without ever breaking the cryptography.
> An anecdote: we find that prefilling Kimi-K3 reasoning with a few tokens of Opus reasoning measurably shifts its response toward Opus’s
> A small memorization analysis showed that specific Claude and GPT reasoning spans are up to ~6 orders of magnitude easier to extract from Kimi-K3 than from the next-closest model.
[1] https://x.com/kotekjedi_ml/status/2087147042888114428?s=42
For example, when there was a paper that came out showing that having model logprobs makes distillation an order of magnitude easier, the closed LLM providers instantly yanked out support for getting the full logprobs at every time step. You get at most top 10 candidates now and I'm sure even that's on the chopping block.
People will use this to argue that a model which has exceeded Opus 4.8 (Kimi K3) somehow got most of its performance through distillation of Opus 4.8.
I still don't buy that distillation was worth more than 3 months of "catch up" time for the chinese labs. Most people who use the word "distillation" to much are revealing their sinophobia.
I can’t fault them too much, as logit based distillation is extremely effective.
Very useful for making smaller models out of bigger open weight models.
You think that you are adding a description for your post when in fact you're simply submitting a regular comment not promoted or distinguished in any way.
It's even worse considering posts without URLs would take the same text from the same input box on the HN submission form and append it under the post title, locking it to the top of the page - so if you've browsed linkless posts you think maybe the submission text is gonna end up just like that, but it doesn't. (Also I think you'll even see URLs like an archive link end up appended just below the submission title for URL posts. Which I guess is a special feature.[1])
Any reason I should not send HN an email requesting clarification of this the submission page?
Edit: quoting https://news.ycombinator.com/submit :
OK, now I finally understand what that means in practice, but it doesn’t imply an entirely separate undistinguished comment will be simultaneously submitted on my behalf.[1]modpowers(?) used to directly append links to URL submissions further confuse the matter: https://news.ycombinator.com/item?id=49243880