Skip to content
HomeBriefFor youTrackersTopicsCompaniesGlossaryNewsletterSecurityHealth & medicineFinance & marketsRoboticsModels & researchPolicy & regulationEnterprise & businessDeveloper tools
Glossary

Preparedness framework (OpenAI)

OpenAI's internal policy for measuring a model's dangerous capabilities before release and deciding which safeguards or restrictions apply.

It scores models in categories such as cybersecurity, biology and chemistry, and AI self-improvement, with thresholds like High and Critical. A High rating requires added mitigations; some categories can block deployment. GPT-6.2 Astra, released in November 2026, was the first externally deployed OpenAI model rated High for AI self-improvement, and both 6.2 models were rated Critical for cybersecurity while remaining on the public API. Anthropic's equivalent is its Responsible Scaling Policy.

← All terms

Get the brief in your inbox

One email every morning: the stories that mattered in AI, each with a one-line "why it matters" for your role. Free, no ads, unsubscribe any time.

Published daily by AI Briefing. We never share your address.
Like this? Get tomorrow's brief by email, free.