VEPA: Verified Evolving Formal Policy for Lifelong LLM Agent Guardrails

Key points

• LLM agents increasingly operate over sensitive resources, such as private documents and mutable system states, which can create serious safety and security risks.1
• Runtime guardrails should therefore make inspectable, reproducible, and deterministic allow or refuse decisions before potentially irreversible agent actions are executed.1
• Existing adaptive guardrails learn from deployment experience, but they usually apply safety policies through LLM inference, which can lead to inconsistent decisions and ambiguous policy interpretations.1
• VEPA represents the guardrail as a formal policy of executable symbolic rules over typed agent events.1

Sources (1)

• [1]VEPA: Verified Evolving Formal Policy for Lifelong LLM Agent Guardrails
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 8, 11:49 PM

  We introduce VEPA, a framework for the verified evolution of formal agent guardrails.

  LLM agents increasingly operate over sensitive resources, such as private documents and mutable system states, which can create serious safety and security risks.

Extractive summary: sentences quoted from the sources.