---
name: ai-application-security
description: Review AI features for prompt injection, data leakage, tool abuse, unsafe output, and excessive agency.
---

# AI Application Security

Map untrusted content, system instructions, retrieval, tools, model outputs, and side effects. Check prompt injection, indirect injection, sensitive-data leakage, insecure tool permissions, output validation, cross-user context, logging, rate limits, and human approval for high-impact actions.