AI Glossary

Frontier safety framework

Also called responsible scaling policy, preparedness framework

What does “Frontier safety framework” mean?

Frontier safety framework: Published policy in which an AI developer commits to test new frontier models for dangerous capabilities, such as helping with cyberattacks or biological weapons, and to apply stronger safeguards or pause deployment if set thresholds are reached. Anthropic, OpenAI and Google DeepMind each publish versions under different names.

Frontier safety framework visual guide: Published policy in which an AI developer commits to test new frontier models for dangerous capabilities, such as helping with cyberattacks or biological weapons, and to apply stronger safeguards or pause deployment if set thresholds are reached. Anthropic, OpenAI and Google DeepMind each publish versions under different names.
Visual guide Open full size (new tab)

What is an example of “Frontier safety framework”?

Under such a framework, a model that crosses a defined risk threshold in testing might ship only with extra security and usage restrictions.

Keep learning

See also

AI safetyTermEvalsTermRed teamingTermFrontier modelTerm

Same topic

GuardrailsTermFairnessTermJailbreakTermExplainabilityTermLeaderboardTerm