Skip to content

EmergenceAI/SafeAgentPP

Repository files navigation

SafeAgentPP

Proving that secrets never leak in multi-agent AI systems — by construction, not testing.

Formal verification of non-interference and safe information flow in multi-agent AI systems using Lean.


Origin

This project was built during the LeanLang for Verified Autonomy Hackathon (April 17–18 + online through May 1, 2026) at the Indian Institute of Science (IISc), Bangalore. Sponsored by Emergence AI Organized by Emergence India Labs in collaboration with IISc Bangalore.


🔥 Key Contributions

SafeAgent++ is a formally verified guardrail framework for multi-agent systems using Lean.

We provide mathematical guarantees for:

  • No Secret Leakage — sensitive data cannot flow to public outputs
  • Non-Interference — secret data does not influence public behavior
  • Safe Multi-Agent Communication — interactions respect trust levels

Unlike heuristic guardrails, our approach provides provable safety guarantees using theorem proving.


🧠 Core Idea

We model information flow using security labels and prove that unsafe flows are impossible using Lean's theorem prover.


⚙️ Build Instructions

lake build

⚠️ Note

This project focuses on formal verification of agentic safety properties using Lean.


🚀 Real-World Impact

Modern AI agent systems can leak sensitive data through indirect interactions.

SafeAgent++ prevents:

  • 🔐 API key leakage (e.g., sending secrets to public agents)
  • 🧠 Prompt injection attacks that expose hidden data
  • 🔄 Unsafe multi-agent communication chains

🧪 Example

A secret API key:

API_KEY_123

cannot be sent to a public agent:

#eval safeInteraction publicAgent secretData  -- false

This behavior is guaranteed by formal proofs in Lean, ensuring correctness across all possible executions.


Acknowledgments

This project was made possible by:

  • Emergence AI — Hackathon sponsor
  • Emergence India Labs — Event organizer and research direction
  • Indian Institute of Science (IISc), Bangalore — Academic partner, hackathon co-design, tutorials, and mentorship

Links

About

No description or website provided.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages