Adversarial Prompt Lab

A hands-on playground for AI agent security

1 follower

A hands-on playground for AI agent security

1 follower

Attack real AI agents with modern adversarial techniques, then turn on defenses and see what changes. Explore prompt injection, MCP poisoning, RAG/memory attacks, excessive agency, zero-click exfiltration, and more — entirely client-side.

Free

Launch tags:Developer Tools•Artificial Intelligence•GitHub

Launch Team

ClarifyThe full-stack AI CRM for modern GTM teams

Promoted

Maker

📌

Hey Product Hunt 👋 I built Adversarial Prompt Lab because I think AI security is easier to understand when you can actually break things. The lab contains 12 hands-on security experiments covering prompt injection, MCP tool poisoning, excessive agency, the lethal trifecta, RAG/memory poisoning, zero-click exfiltration and more. The workflow is deliberately simple: Attack → Observe → Defend → Attack again. Everything runs client-side and you bring your own API key. No backend, no telemetry, and the attacks use benign canaries rather than real secrets. I'd love feedback from security researchers, AI engineers, red teamers, and anyone building agents: What attack or defense should I add next? Try breaking it. 🧠 — Sonu / AI Anytime

Report

2mo ago

Reviews

No reviews yetBe the first to leave a review for Adversarial Prompt Lab