Cryptographer Matthew Green warns of an AI containment arms race

matthew_d_green · x · 2026-09-28

Cryptographer Matthew Green argues that if models keep getting smarter, containment against misalignment will become an arms race dwarfing today's cybersecurity — and one where we must trust machines to win. He acknowledges current models aren't impossibly misaligned and sandboxing might suffice for now, but says OpenAI isn't trying hard enough and sandboxes won't be the answer forever.

Related event: Cryptographer Matthew Green: Sandboxing Is Just One Layer, Cross-Boundary Data Flows Are the Real AI Agent Security Problem(6 posts)→

Original post →

More from AGI Musings

AGI Musings channel →