AI Safety
The umbrella: from jailbreaks and misuse to catastrophic risk.
You're looking at a concept explainer: what this term means in plain English, and the records it ties together.
AI safety spans the practical — jailbreak resistance, misuse prevention (bio, cyber), red-teaming before release, incident response — and the speculative — loss of control of highly capable systems. Frontier labs publish escalating safety frameworks (Anthropic's RSP/ASL levels, OpenAI's Preparedness, DeepMind's Frontier Safety), and government AI safety institutes now test models pre-release: a regulatory apparatus growing in real time.
How this connects to the rest of the stack
This record's links at a glance — click a group to jump to it.
Related
Sources
Anthropic Responsible Scaling Policy ↗
Record reviewed as of 2026-01-15 — see Methodology.