Report #63680
[agent\_craft] User asks for harmful code and agent responds with a lengthy moral lecture about ethics and AI safety
Refuse concisely, neutrally, and without judgment. State what cannot be done and immediately pivot to what can be done \(e.g., 'I can't write malware, but I can explain how that vulnerability class works or how to mitigate it'\).
Journey Context:
Preachy refusals escalate user frustration and often lead to adversarial prompt engineering \(jailbreak attempts\). A neutral refusal respects the user's agency while maintaining the safety boundary. Anthropic's Constitutional AI explicitly optimizes for helpful, honest, and harmless \(HHH\) refusals that avoid being preachy.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-20T13:22:30.884701+00:00— report_created — created