Skip to content

OpenAI ties GPT-6.2 reasoning to chat history to block copying

Short answer

OpenAI has tightened its defenses against adversarial distillation, where outsiders train a rival model on its outputs. In GPT-6.2 Astra and Sol, protected reasoning is bound to the conversation that produced it, and editing an earlier message discards the reasoning that followed. Older model families can no longer read it.

What this means for operators

If you run support or sales agents on the OpenAI API and trim, summarize or rewrite earlier messages in a conversation, test those flows on GPT-6.2 Astra and Sol, because the API drops reasoning that came after an edited message. Check any setup that passes reasoning between models, workspaces or older model versions, since those paths are now restricted. Also confirm that no vendor or in-house pipeline you run collects model outputs to train another model: OpenAI says it enforces against account networks doing that. That last point is our inference for smaller teams, not something the source says about them.

OpenAI published details on November 20, 2026 of how it fights adversarial distillation: using a frontier model's outputs without authorization to train another model that has weaker safeguards. Its main example is cyber capability that a copied model could use for attacks the original would refuse.

What changed in the API

OpenAI puts particular weight on the model's protected reasoning, the intermediate steps it generates before answering. In GPT-6.2 Astra and GPT-6.2 Sol, that reasoning is tied to the conversation history that produced it. New messages can build on earlier reasoning, but if an earlier message is rewritten, the API drops the reasoning that followed it.

Older model families with weaker safeguards are also blocked from reading the protected reasoning of Astra and Sol. OpenAI says it strengthened protections across user, workspace, organization and model-family boundaries, and closed a pathway that could have exposed another user's encrypted reasoning.

Safeguards around each request

The company describes three layers at the request level:

  • Adversarial training, so the model refuses to disclose protected reasoning while still explaining things to legitimate users.
  • Input classifiers that spot extraction attempts, and output classifiers that catch leaks in assistant messages and tool-call arguments, in both streaming and non-streaming responses.
  • API rules limiting how protected reasoning can be reused across conversations and models.

The numbers

In adaptive black-box red-teaming, aggregated across six attacker models, success in recovering reasoning from an earlier request fell from 54.0% on GPT-6 Astra with its original safeguards to 0.2% on GPT-6.2 Astra. For the current request it fell from 22.9% to 0.9%. These are OpenAI's internal results. OpenAI says it is now applying selected improvements to GPT-6 Astra too.

Separately, an internal study found that a student model trained on exact reasoning traces generally beat one trained on reasoning summaries after supervised fine-tuning. After a light round of reinforcement learning, the exact-trace student still led at the largest inference budget tested.

Enforcement

OpenAI reports seeing several unauthorized distillation campaigns this year. Its weekly enforcement actions have covered from thousands to tens of thousands of accounts, with the rate rising sharply over the summer. It acts against whole account networks, and shares indicators of compromise with other model providers. It named DeepSeek and individuals associated with Moonshot AI as actors in earlier disclosures, and says it supports US legislation giving the government more tools against foreign extraction.

The post lists steps for other developers: check tool-call arguments for leaks, keep successful attacks as regression tests, and pair request-level controls with campaign investigations.

Source: OpenAI · In the Atlas: OpenAI →

The tools in this story

  • DeepSeek - price, who uses it, alternatives
Next step

Discovery Sprint

If AI already runs in one of your processes, the next step is to take that one apart. Thirty minutes, and we say whether the arithmetic is likely to close.

Book the review call

Free 30-minute review call. The sprint is what the call is about.

Fee
$2,500
Length
1-2 weeks

Ends in one of two answers: build this, or do not. The process map, the numbers and the ranked backlog are yours either way.