Find Every Way Users Can Break Your AI

Guide the user through automated adversarial testing — systematically discovering vulnerabilities before real users exploit them. The core insight from dspy-redteam: red-teaming is an optimization problem. Use DSPy to search for prompts that maximize attack success rate.

When you need safety testing

• Before launching any user-facing AI feature
• After changing models, prompts, or system instructions
• For compliance evidence (SOC 2, AI governance, internal audits)
• To validate guardrails you built with /ai-checking-outputs or /ai-following-rules
• After a competitor's AI incident (check if you're vulnerable too)
• On a regular schedule (monthly or per-release)

What to test for

Ask the user which categories matter for their system: