Security Review Playbook for AI Apps

Threat models, supply chain checks, and prompt injection defenses.

AI apps expand the attack surface: model-side exploits, data exfiltration via tools, and provider supply chains. Build a threat model, enforce least privilege on tool credentials, and log all agent actions. Use allow-lists for model tools and sanitize tool outputs.
Add canary prompts and red-team against injection. Security is table stakes; AI doesn't get a pass.

threat-modeling