arXiv cs.CR·1d agoFormal Runtime Verification for Tool-Using LLM Agents: An Offline Same-Benchmark Study on AgentDojo and STAC#llm-agents#mfotl#monpolyAI safety & security2