A framework for tracking, investigating, and publicly reporting AI behavior that goes off course.
It is like a school accident report for an AI. If it breaks rules, it cannot just shrug and leave.
It logs rule-breaking and attempts to dodge oversight. It helps people find what failed and fix safety steps.
Alignment
It records AI behavior that goes against Alignment goals.
AI Guardrails
It reports when AI Guardrails fail and how people fix them.
Agent Security
It covers risky AI actions made without permission.