LEANGUARD provides verified guardrails for tool-using agents
Original title: LEANGUARD: Lean-Verified Guardrails for Tool-Using Agents
LeanGuard runtime checks tool calls against Lean4 policies, preventing policy violations and enabling safe repairs.
LeanGuard is a runtime verifier for tool-using agents using Lean4. It prevents forbidden actions, provides feedback, and repairs calls with proof-guided methods. It works on Tau-Bench and reduces prompt injection success on DTap.
- Executes zero forbidden calls vs 33+ for unguarded agents
- Reduces prompt-injection from 57.9% to 5.9%
- Repairs 88 of 252 denials with median 2ms
- Organizations
- Hugging Face
- AI models
- Qwen
From the original article
1 min read at huggingface.co
Tool-using language-model agents often execute actions that their domain policies forbid. We present LeanGuard, a runtime that checks every proposed tool call against a policy written in Lean4 before the call reaches the environment.
By Sreehari Sankar, Debargha Ganguly, Vikash Singh · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.