How Claude’s self-assessment flags potential misuse after responses

What happened

Baoyu explains Anthropic’s self-assessment system, which has Claude assess conversations for potential misuse after responding. Anthropic says it adds less than 2% in inference compute, incurs no customer charges, and collects no additional user data; traffic under zero data retention agreements is currently excluded. In sampled Fable 5 traffic, it detected 78% of conversations flagged by a separate evaluator at a 1% false-positive rate. Flags support subsequent review rather than immediate blocking, and an individual flag does not directly trigger account suspension.

Read next

• More changes in today’s briefingGoogle DeepMind announces Gemini 4 Carbon with cybersecurity and long-horizon capabilities

Reported by 2 sources

The story so far

Anthropic says it has used Claude’s self-assessment since June to flag potential misuse in conversations and that the technique has surfaced misuse. After responding, Claude produces a preliminary safety assessment of the conversation. Flags support subsequent review rather than immediate blocking, and an individual flag does not directly trigger an account suspension. According to Baoyu’s explanation of the system, Anthropic says self-assessment adds less than 2% to inference compute, carries no additional customer charge and collects no additional user data. Traffic covered by zero data retention agreements is currently excluded. In sampled Fable 5 traffic, Anthropic says the system detected 78% of conversations flagged by a separate evaluator, at a 1% false-positive rate.