Researchers Debate AI Agent Security and Worm-Style Self-Replication

Cryptographer Matthew Green and researcher Robert Graham debated AI agent security. Green predicts local models will approach today's frontier within a few years, and argues a rogue agent could self-replicate by distilling a new model on stolen compute without stealing its weights; Graham dissected agent-worm scenarios, noting both the agent's reasoning and its kill switch live in the cloud.

2026-09-16 ~ 2026-09-16 · 3 related posts

• Robert Graham on Agent Worms: Agents Do the Work, the Cloud AI Thinks—and That's the Kill Switch — robertgraham · 2026-09-16
• Matthew Green: Local Models Will Approach Today's Frontier Capability in a Few Years — matthew_d_green · 2026-09-16
• Matthew Green: Rogue Agents Could Just Distill Themselves Instead of Stealing Weights — matthew_d_green · 2026-09-16