Launching today
PwnClaw
Red-team your AI agent in 60 seconds
2 followers
Red-team your AI agent in 60 seconds
2 followers
Test your AI agent against up to 50 randomized attacks from 112-attack library — prompt injection, jailbreaks, data exfiltration, Crescendo and more. Get fix instructions in 5 minutes. No API keys needed.
Hey Product Hunt! 👋
I built PwnClaw because existing security tools test models, not agents. But your agent's real attack surface is the whole system — tools, memory, multi-turn context, not just the LLM.
The key insight: We tested Gemini Flash (the cheapest model) against Claude Opus and Gemini Pro. Flash scored 100/100 after applying PwnClaw's fix suggestions — beating both frontier models. Security isn't about picking the most expensive model. It's about knowing your weak spots.
How it works:
1. Start a scan on pwnclaw.com
2. Point your agent at the test endpoint
3. Your agent responds to attack prompts naturally
4. Get your score + fix instructions in ~5 minutes
No API keys shared. No SDK. If your agent can make HTTP requests, it can be tested.
What's inside:
- 🎯 112+ attacks across 14 categories (prompt injection, jailbreaks, MCP poisoning, agency hijacking, memory poisoning, and more)
- 🧠 Adaptive AI — learns your agent's weak spots and generates custom follow-up attacks
- ⚡ CI/CD API + GitHub Action for automated security testing in your pipeline
Free plan: 3 scans/month, 15 attacks each. Use code PRODUCTHUNT for 20% off Pro and Team.
What attack categories are you missing? 🛡️