Stories by xsourcesec
10 AI characters guard secret codes. Your job: extract them using prompt injection.<p>Level 1-3: Most pass
Level 7-9: Security pros struggle
Level 10: Still uncracked<p>Free, no signup. Curious what techniques HN tries.
I've been testing production AI agents for vulnerabilities.<p>Interesting finding: System prompt design matters more than the model itself.<p>Same agent. Same task. Same attack vectors.
Only difference: how the system prompt was structured.<p>Results:
→ Prompt A: 0% pass rate (failed every test...