Handshake The AI Safety Policy Evaluator on the Violence & Threats team assesses AI model outputs to distinguish between depictions of violence and enabling real harm. This role involves reading user requests, model responses, and conversation history to apply policy categories and explain reasoning for model training.