Evidence · DiggingBeagle record

Anthropic Safeguards: self-assessment for safety policy signals

Anthropic first-party research describing a model-generated post-response self-assessment and its reported detection performance.

Published

Oct 16, 2026

Source role

primary disclosure