AI Glossary

AI safety

What does “AI safety” mean?

AI safety: Broad field concerned with preventing AI systems from causing harm, whether through accidents, misuse by people, or behaviour nobody intended. It spans technical research, testing, deployment policies and governance, from everyday chatbot mistakes to risks from far more capable future models.

AI safety visual guide: Broad field concerned with preventing AI systems from causing harm, whether through accidents, misuse by people, or behaviour nobody intended. It spans technical research, testing, deployment policies and governance, from everyday chatbot mistakes to risks from far more capable future models.
Visual guide Open full size (new tab)

What is an example of “AI safety”?

Testing whether a model can give meaningful help toward building weapons, before releasing it, is a typical AI safety activity.

Keep learning

See also

AlignmentTermRed teamingTermFrontier safety frameworkTermResponsible AITerm

Same topic

Mechanistic interpretabilityTermAI regulationTermArena rankingsTermAI biasTermBenchmarkTerm