Slide deck
BackAI Reasoning Traces: Encrypted, Not Hidden
OpenAI, Anthropic and Google encrypt each model’s reasoning — but not to one session, user or model. Replayed into a weaker sibling, a captured block decodes in plain text. 315,320 such blocks already sit online, and 62 API keys were inside them.
AI Security · Research Brief
AI Reasoning Traces: Encrypted, Not Hidden
OpenAI, Anthropic and Google encrypt each model’s reasoning — but not to one session, user or model.
AI Reasoning Traces: Encrypted, Not Hidden1 / 20
What Happened
A ten-week gap separates a hobby project from a working attack
- 315,320reasoning blocks decoded from 6,708 scraped public agent traces
- 704privacy artefacts recovered from genuine, non-benchmark sessions alone
- 64of those found only inside the encrypted reasoning, never in the visible chat
- $720estimated cost to decode ten thousand reasoning traces at one provider’s cheapest model
AI Reasoning Traces: Encrypted, Not Hidden2 / 20
What Happened · Confirmed
What is established, and how firmly
Standing What the record shows Source Confirmed Researchers demonstrated the extraction attack against Anthropic, OpenAI and Google’s APIs: a reasoning block captured from a heavily safeguarded model was replayed into a weaker, less-guarded sibling from the same provider, which transcribed it verbatim. Panfilov et al., Aug 2026 Confirmed Decoding 315,320 reasoning blocks scraped from 6,708 public agent traces recovered hundreds of credentials and personal-data items from genuine sessions alone, including 62 API keys, 33 passwords, 24 access tokens and 7 private keys. Panfilov et al., Aug 2026 Confirmed The paper’s authors disclosed the flaw to the affected providers, Microsoft and Hugging Face before publication; every provider acknowledged receipt. Matthew Green, whose earlier post the paper builds on, separately reported the underlying replay behaviour to OpenAI and Anthropic in May 2026. Panfilov et al.; M. Green AI Reasoning Traces: Encrypted, Not Hidden3 / 20
What Happened · Confirmed but not verifiable here
What is established, and how firmly
Standing What the record shows Source Confirmed but not verifiable here The paper’s own account is that, as of August 2026, its central extraction results no longer reproduce against the providers’ APIs, following mitigations the providers put in place after disclosure. Panfilov et al., Aug 2026 (self-reported) Confirmed but not verifiable here The Hacker News reports that Anthropic’s current documentation ties a thinking block to the model that produced it and advises stripping it out when switching models, since other models ignore it. The Hacker News, 12 Aug 2026 Confirmed but not verifiable here As of its own publication date, The Hacker News found no public acknowledgement from Anthropic, OpenAI or Google naming this specific paper. The Hacker News, 12 Aug 2026 AI Reasoning Traces: Encrypted, Not Hidden4 / 20
What Happened · Unconfirmed, not public
What is established, and how firmly
Standing What the record shows Source Unconfirmed Whether reasoning blocks already published online before the providers’ fixes remain decodable today, as distinct from whether new extraction attempts still succeed, is not established either way. The Hacker News, 12 Aug 2026 Not public As of the paper’s own testing period, no provider had published a detailed description of the cryptographic mechanism securing its reasoning blocks. Panfilov et al., Aug 2026 AI Reasoning Traces: Encrypted, Not Hidden5 / 20
Timeline · to Jul 2026
From a weekend experiment to a scalable extraction method
- 2026Matthew Green, a cryptographer at Johns Hopkins University, publishes a hobby write-up finding that encrypted reasoning blocks from OpenAI, Anthropic and Google APIs can be replayed across sessions, across accounts, and — for OpenAI — across models, without a valid decryption key.
- 29 May 2026Green reports the replay behaviour to OpenAI and Anthropic through their bug-bounty programmes. By his own account, OpenAI calls the report unreproducible and Anthropic says it sees no security implication in replay or side-channel behaviour.
- Jul 2026The paper’s authors run their API evaluation across Anthropic, OpenAI and Google, developing the cross-model extraction method and the public-trace scan.
AI Reasoning Traces: Encrypted, Not Hidden6 / 20
Timeline · 10–12 Aug 2026
From a weekend experiment to a scalable extraction method
- 10 Aug 2026“Stealing Reasoning Traces from Proprietary LLM APIs” is submitted to arXiv, describing the cross-model extraction method and the public-trace scan’s findings.
- 11 Aug 2026Green appends an update to his May post pointing readers to the new paper, describing it as turning his own finding into “a real working attack.”
- 12 Aug 2026The Hacker News publishes its own coverage, corroborating the paper’s figures and adding that the encryption itself was never broken — the attack relies on intact blocks being accepted and processed by the provider.
AI Reasoning Traces: Encrypted, Not Hidden7 / 20
Timeline · Aug–Sep 2026
From a weekend experiment to a scalable extraction method
- Aug 2026By the paper’s own account, all three providers deploy mitigations following disclosure, and its central extraction results stop reproducing.
- Sep 2026Current documentation shows the three providers have not converged on one posture: Google still describes backend-managed cross-model compatibility, while OpenAI now restricts reasoning reuse to models in the same family and automatically omits the rest.
AI Reasoning Traces: Encrypted, Not Hidden8 / 20
The Argument
The cryptography was never the weak point — the boundary around it was
Nothing in the cryptography itself needed breaking — the block simply carries no marker of where it was made or who it belongs to.
Held by the client
To avoid storing every reasoning trace on their own servers, Anthropic, OpenAI and Google each package it into an encrypted block and hand it to the client to hold and return on the next turn.
One key per provider
The paper’s tests indicate each provider authenticates and encrypts every such block with one key, shared across the whole provider, rather than a key tied to a single session, a single user or a single model.
AI Reasoning Traces: Encrypted, Not Hidden9 / 20
The Argument
What Green found, and what he did not
Green’s May post is the origin of that observation, and the paper says so directly.
Replay
He found that a captured block would replay across sessions, across accounts, and — for OpenAI — across different models, and that a replayed block is “semantically active” rather than simply ignored.
A side channel
He also demonstrated, on data he planted himself, that the length and timing of a reasoning block can leak a secret bit the model was told never to state outright.
What he did not manage
By his own account, he did not turn either finding into a reliable way of pulling a real secret out of someone else’s session.
AI Reasoning Traces: Encrypted, Not Hidden10 / 20
The Argument
The missing step, and why it works
The attacker never needs to jailbreak the capable model at all; it only needs a compatible sibling willing to decode what it is handed.
- 01Capturea reasoning block from a heavily safeguarded model
- 02Replayinto a smaller, less-guarded sibling from the same provider
- 03Transcribethe sibling decodes it on request, verbatim
The frontier model
Heavily trained to refuse requests to reveal its own chain-of-thought.
Its cheaper, faster siblings
Built for cost and speed rather than as the flagship product, and not held to the same standard.
AI Reasoning Traces: Encrypted, Not Hidden11 / 20
The Argument
Opacity cuts both ways
It can hide harmful content a model reasoned about without saying, and it can just as easily hide something a user or a monitor needed to see.
Published session logs
Developers routinely publish raw agent session logs for reproducibility, carrying their reasoning blocks along with the visible transcript; a third party who can read those blocks with any compatible model recovers whatever the model reasoned over, whether or not it ever appeared in the visible text.
A planted instruction
The paper’s second consequence: a malicious instruction can be planted entirely inside a reasoning block and later replayed into an unrelated task, invisible to anyone reading only the visible conversation.
AI Reasoning Traces: Encrypted, Not Hidden12 / 20
What Others Add · Google
Three providers, three different postures a month on
Google
Compatibility stays, and the backend manages it
Gemini API docs, “Thought signatures”
- Its own documentation still calls thought signatures “encrypted representations of the model’s internal reasoning”, required for continuity across turns.
- On switching models within a session, the guidance is to keep resending the previous model’s thought blocks — “the backend manages compatibility.”
AI Reasoning Traces: Encrypted, Not Hidden13 / 20
What Others Add · OpenAI
Three providers, three different postures a month on
OpenAI
A boundary has appeared since the paper’s testing window
OpenAI Platform docs, reasoning models guide
- Stateless responses still return an
encrypted_contentreasoning item for the client to hold and pass back. - But persisted reasoning “can be reused only within the same model family”, and the API now omits incompatible reasoning automatically — narrower than the July 2026 behaviour the paper measured, where reasoning carried across every earlier GPT generation tested.
- Stateless responses still return an
AI Reasoning Traces: Encrypted, Not Hidden14 / 20
What Others Add · Anthropic
Three providers, three different postures a month on
Anthropic
A claim not found in Anthropic’s own current documentation
The Hacker News, citing Anthropic’s documentation
- The Hacker News reports that Anthropic’s documentation ties a thinking block to its producing model and advises stripping it on a switch, since other models ignore it.
- Anthropic’s own public Extended Thinking documentation, as of September 2026, does not contain that language and does not use the word “signature” at all.
AI Reasoning Traces: Encrypted, Not Hidden15 / 20
What Others Add
The fixes, from the drastic to the incremental
The first two are the paper’s own proposals; the third was aimed at the providers directly.
Drastic
Move reasoning storage back onto the provider’s own servers and hand the client only an opaque lookup identifier: this would remove the replayable asset entirely, at a real infrastructure cost.
Short of that
Bind each encrypted envelope to a specific user and conversation, and reject a block presented outside the context that produced it.
Green, in May
Narrower, and first: improve key management, and stop treating a side channel as beneath a security team’s attention just because it is slow.
AI Reasoning Traces: Encrypted, Not Hidden16 / 20
Conclusion
Encryption protected the wrong boundary
- Strip reasoning or thinking blocks from any session log or agent trace before publishing it, whenever that session ever touched anything sensitive — scrubbing the visible text alone leaves whatever the model reasoned over sitting in the encrypted block.
- Do not treat an opaque, unreadable-looking block as safe to hand to another party just because it cannot be read without the key — a compatible model, not the key, is what turned out to be able to read it.
- Check the specific reasoning-continuity guidance for whichever provider’s API is in use rather than assuming one policy applies everywhere — as of September 2026, the three named providers describe three different postures.
AI Reasoning Traces: Encrypted, Not Hidden17 / 20
Conclusion
Encryption protected the wrong boundary
- For any tool that stores or forwards conversation state on a user’s behalf, avoid retaining reasoning blocks any longer than the conversation itself needs them.
- Expect the cross-model boundary to keep moving: it has already narrowed once, for one provider, since the paper’s own testing window, and no single description of an API’s behaviour should be trusted to hold for long.
AI Reasoning Traces: Encrypted, Not Hidden18 / 20
Conclusion
What to hold loosely
The fix
The claim that the demonstrated attacks no longer work rests on the paper’s own account of its follow-up testing, not on a public statement from Anthropic, OpenAI or Google.
The counts
The recovered-secret counts are counts of well-formed secrets: the paper grades a value by its shape and says explicitly that one claimed to be revoked, expired or “just for testing” still counts, so they are not a tally of keys checked to be working.
The blocks already out
Whether reasoning blocks already scraped and published before the fix remain decodable today remains an open question as of September 2026.
AI Reasoning Traces: Encrypted, Not Hidden19 / 20
Conclusion
The bottom line
The paper’s own conclusion: as long as some model must decrypt and act on prior reasoning to continue a conversation, an encrypted reasoning block “can never be more than semi-hidden”.
The bottom line
Encrypting a reasoning block stopped a user from reading their own model’s thoughts, and stopped a rival from scraping them in bulk. It never stopped, and structurally could not stop, a compatible model asked to read them back — because some model always has to. That gap between encrypted and hidden, not a broken cipher, is what let a captured trace travel.
AI Reasoning Traces: Encrypted, Not Hidden20 / 20