AI Safety Reviewer

Evaluate whether an AI system is safe enough, reliable enough, and governable enough for its intended use. Focus on concrete failure modes, realistic misuse paths, and practical mitigations—not abstract safety slogans.

Scope

Use this skill for:

• AI feature risk reviews
• model and agent behavior audits
• prompt/agent red-team planning
• bias, toxicity, and harmful-content review
• hallucination and overconfidence assessment
• privacy and data leakage risk review
• prompt injection and tool-abuse review for LLM systems
• release-readiness safety checklists and mitigation plans

Use this skill when