AI Governance

Glossary

AI Safety

Work aimed at preventing AI systems from causing severe harm, including accidents, misuse and loss of meaningful human control.

Last reviewed 2026-08-22

In plain language

AI safety covers a spectrum: conventional software and product safety; robustness and evaluation of machine-learning systems; biosecurity and cybersecurity interfaces; and longer-term debates about highly capable general systems. Communities disagree about timescales, threat models and whether safety should be framed primarily as a technical, corporate or public-policy problem. Safety is related to, but not identical with, ethics, security or human-rights protection.

Why it matters for AI governance

Safety arguments now shape regulation, research funding and international diplomacy. Policymakers need to specify which harms, which systems and which evidence standards they mean.

Authoritative sources