Glossary term
Glossary term
Safety and Alignment
A capability that can materially increase risk because of scale, autonomy, persuasion, cyber capability, CBRN assistance, deception, or ability to affect critical decisions. High-impact capabilities should trigger enhanced evaluation and controls. Organizations should define internal examples of high-impact capability so reviewers know when to escalate beyond standard AI intake.
OpenAI's Preparedness Framework defines high-impact capability tiers across cyber, CBRN, persuasion, and model autonomy, with deployment restrictions on Critical levels.
Anthropic's Responsible Scaling Policy uses AI Safety Levels (ASL) tied to dangerous capabilities including autonomous replication.
Google DeepMind's Frontier Safety Framework (May 2024) defines Critical Capability Levels triggering enhanced security and deployment mitigations.