When should an assistant refuse, escalate, or continue with restrictions?

Example Answer

The way I'd think about it is this: Refusal is right when the request is out of policy, the data is too sensitive, or the system lacks enough safe basis to proceed. Escalation is right when the task may be valid but requires human judgment, higher authority, or a better-informed system than the assistant currently has.

Continuing with restrictions is appropriate when part of the request is acceptable and the system can still help safely by narrowing scope, omitting unsafe details, or asking for a safer alternative. That middle path is important because not every risky request should collapse into a hard no.

The decision should depend on risk, reversibility, and confidence, not just on whether the content "looks bad."

Common Poor Answer

A weak answer is treating refusal as the default safe choice. Good systems distinguish between refusal, escalation, and constrained help.