Hey Product Hunt! I'm Javeria, and I built Raven because I kept seeing AI chatbots break character or leak their system prompts with fairly simple attacks, and there wasn't an easy way to actually test for that before shipping. Raven lets you paste your chatbot's system prompt and runs it through real prompt injection, jailbreak, and information-leakage attacks across six OWASP LLM Top 10 categories. You get a risk score, a breakdown by attack type, and actionable fixes, no API key to your live bot required, no signup, completely free. Under the hood it uses a hybrid detection system: fast pattern matching for clear-cut cases, plus a small local classifier for ambiguous refusals, so it's not burning a second LLM call just to judge the first one. Would genuinely love for you to try it on your own system prompts and tell me what breaks or what false positives you hit, this is early and I'm actively improving the attack library. Happy to answer anything about the detection approach in the comments!

Raven

AI system prompt vulnerability scanner

AI system prompt vulnerability scanner