Gemini: Response Blocked by Safety Filter

Gemini's safety system evaluates both user input and model output against four harm categories: hate speech, dangerous content, sexual content, and harassment. When the model's generated response scores above the safety threshold for any category, Gemini blocks the response before returning it — even when the user's input was benign. In the API, this manifests as a finish_reason of SAFETY and a None content field, causing response.text to raise an AttributeError.