HOW WE INCLUDE AND CLASSIFY AI SAFETY & CYBER SECURITY EVENTS

Return to the AI — Safety & Cyber Security Tracker

INCLUSION STANDARD

An event belongs in this tracker when it provides credible evidence that materially changes at least one of the following:

• The demonstrated capability of an AI system to act autonomously, deceptively or beyond its intended boundary.
• The ability of AI to increase the scale, speed, adaptability or accessibility of cyberattack.
• The ability of institutions to detect, contain or govern those capabilities.
• The beliefs and actions of people with direct knowledge of frontier AI development.
• The rules, policies or competitive pressures shaping safety decisions.

The tracker distinguishes verified events, reported claims, expert assessments and forecasts. A warning is evidence that a qualified person holds a concern; it is not automatically proof that the predicted outcome will occur.