agent-exploit

Live demonstrations of prompt injection vulnerabilities in agentic AI systems

install

npm install

usage

// Run a specific exploit demo
node demos/data-exfiltration.js

// Test tool hijacking vulnerability
node demos/tool-hijacking.js

// Run prompt injection chain
node demos/injection-chain.js

// Start interactive demo server
npm start

Each demo shows a different attack vector:

• data-exfiltration.js - Extract sensitive data through hidden prompts
• tool-hijacking.js - Force agents to call unintended functions
• injection-chain.js - Chain multiple injections across agent interactions
• goal-hijacking.js - Redirect agent objectives mid-task

Configure target LLM providers in config.json.

license

MIT