arXiv cs.CR·10d agoThe Verifiable Action Card: Trustworthy Human-in-the-Loop Control for Secure Autonomous Agents#agentic-browsers#agent-security#confused-deputyAI safety & security