Governing AI Agents: Alan Chan of GovAI on Monitoring, Agent IDs, Rogue Agents & Preparing for RSI

Frontier labs still lack synchronous monitoring, agent identifiers, and enforceable pacing commitments needed to prevent loss-of-control as AI R&D automation accelerates.

• Monitoring gaps remain: Centralized deployments at OpenAI/Anthropic use input-output and contextual classifiers, but synchronous real-time blocking is still inconsistent — the UK AISI incident lacked it due to time pressure, not technical infeasibility.
• Agent identity is unsolved: No standard exists for agents to identify themselves online; most "pose as humans," creating what Chan calls a detection-evasion arms race as platforms like Amazon block bots like Meta's Muse.

Themes, sentiment and reasons are AI-written paraphrases, checked so they never quote an episode. They may contain errors. Not investment advice. Episodes remain the property of their creators.