Saturday, 21 November 2026, story 3 of 7
OpenAI says reasoning extraction from GPT-6.2 Astra fell to 0.2%
In OpenAI's internal red-team test, extraction from an earlier request fell from 54.0% on GPT-6 Astra; the current-request rate is 0.9%.

What happened
OpenAI says it has sharply cut the success of attacks that extract protected reasoning from its newest models. In its internal adaptive red-team tests, attacks recovering reasoning from an earlier request succeeded 54.0% of the time against GPT-6 Astra and 0.2% against GPT-6.2 Astra. For the current request, the rate fell from 22.9% to 0.9%.
The safeguards cover GPT-6.2 Astra and GPT-6.2 Sol. OpenAI says weekly enforcement against distillation campaigns has ranged from thousands to tens of thousands of accounts. It is applying selected improvements to GPT-6 Astra as well.
What it means
The numbers are OpenAI's own, from automated attackers working black-box, so they show progress, not proof that extraction is closed. The 0.9% on current requests is small but not zero.
For builders, the API changes matter more than the percentages. OpenAI says reasoning is now tied to the conversation history that produced it, and rewriting an earlier message drops the reasoning that followed. Apps that edit history may therefore lose reasoning continuity. Older model families are also blocked from reading GPT-6.2 reasoning, so pipelines that hand work between model generations may behave differently. The source does not say how such failures appear.
Teams that train on API outputs should note that enforcement targets networks of accounts. OpenAI's cyber study is mixed: exact traces helped most only at the largest inference budget tested.
What to do: If your app rewrites conversation history or passes reasoning between model families, test it against GPT-6.2 Astra and Sol before relying on continuity.
Sources
Filed under Safety and security. Published .
"What it means" is analysis, not reporting. Found an error? Write to daily@vibitely.com.