Anthropic introduced a technique called self-assessment that lets Claude flag potential misuse immediately after responding to users. The company began deploying the feature in June, noting that it catches harms missed by traditional safety classifiers.
FireBrief
FireBrief