Definition
Plain language
OpenAI's written rulebook for deciding when a model is too dangerous to release without extra safeguards.
As stated in the literature
OpenAI's risk policy defining tracked capability categories (e.g. cyber, biological) with graded thresholds up to Critical, tied to required mitigations and deployment decisions.
Why it matters: Written-down rules turn release decisions into something outsiders can point at and hold a company to, instead of a judgment call made behind closed doors.
For example, if evaluations show a model has reached the highest cyber risk tier, the framework specifies what safeguards must be in place before it can be deployed.
Heard on the show
“Astra is the first model OpenAI has rated Critical for cyber capability under its Preparedness Framework.”Episode 259 — GPT-6 Astra Behaves Better, And OpenAI Can Read It Less