Governing AI Agents at the Hardware Boundary
summary
Shows a concrete stack: the open‑source Agent Governance Toolkit plus a TEE-backed CMCP runtime that runs policies inside enclaves and emits verifiable trace logs, a proposed trace standard for auditor-friendly tamper‑evidence, and an agent manifest that hashes prompts, models, policies and identity. Demonstrations run a local MCP, show policy changes causing agent invalidation, and produce attestations and hashes so operators can prove which policy and hardware produced each decision; AGT 4.1 starts emitting trace logs.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:25Software vs hardware governance
- 3:11Software-only limits
- 7:06Governance Toolkit overview
- 8:50Confidential MCP runtime
- 10:55Trace standard for proof
- 13:03Agent manifest deployment
- 15:53Attestation demos
- 21:22Performance overhead
- 22:47Summary
related talks
33mGoverning the Ungovernable: Security and Compliance for AI Agents in Open Source Projec...
50mMCP at Scale: Governing AI Agent APIs When Autonomous Systems Are Your Consume…
40mIn AI We Trust? Securing the Future, One Agent at a Time
40mAgents in Action: Advancing Open Source AI Missions
33mWhat Did My Agent Do? Observability and Accountability for AI Agents
36mPutting AI Agents to Work: A New Era of Open Source Connectivity with MCP
08magentctl: A CLI for Managing AI Agents & MCP Servers Like Infra
22mAccountable AI agents