Skip to main content

AI news

OpenAI details safeguards against protected-reasoning extraction

OpenAI described updated classifiers and API controls intended to prevent unauthorized extraction of its models’ protected reasoning. It says GPT-6.2 Astra and Sol bind protected reasoning to the conversation history and block older model families from reading it. OpenAI reported a drop in one internal adaptive test from 54.0% attack success on GPT-6 Astra with original safeguards to 0.2% on GPT-6.2 Astra with updated safeguards; this is a vendor-run evaluation, not a guarantee against attacks.

Primary source · OpenAI · November 20, 2026

Opens OpenAI in a new tab. Read it there before you rely on the summary above.

Read the original

Free, email only

Unlock the full brief free.

  • What to check before you trust this story, written down
  • Each verified source, with why it matters and who published it
  • A note whenever a source has been withdrawn
  • Thirty days of stories to browse, not seven

Continuing subscribes the email on that account to Work Worth Doing, my newsletter, the same as the form below, and unlocks every free tool in this browser. We only receive your email address. Unsubscribe in one click.

This is not an account: there is no password, and the unlock is a cookie in this browser. You also join Work Worth Doing, Demetri Panici's newsletter, every week: what I built and what changed in AI. We'll email you a link to confirm, and you can unsubscribe in one click. The same signup unlocks every free tool on the site. How your email is handled.

Already subscribed? Enter the same email to unlock this browser. You won't be signed up twice.