Gemini: Request Refused by Policy

Unlike the blocked safety response (a technical filter), this is a model-level refusal where Gemini chooses not to respond based on its training and policy guidelines. This applies to requests for medical diagnoses, legal advice, instructions for dangerous activities, impersonating real people, and certain creative content. The refusal appears as an in-conversation message rather than an error overlay, and there is no finish_reason: SAFETY in the API — the model generates the refusal as its actual response.