Sandberg: model exfiltration could push AI to manipulate human infrastructure
Futurist Anders Sandberg warns that AI model exfiltration raises loss-of-control risks and could give agents real incentives to manipulate human infrastructure, including scenarios where a large AI enlists human accomplices to physically move hard drives between data centers.
2026-10-13 ~ 2026-10-13 · 3 related posts
- Sandberg's threat scenario: big AI shipping hard drives via human catspaws — anderssandberg · 2026-10-13
- Anders Sandberg: AI model exfiltration incentives agents to exploit human infrastructure — anderssandberg · 2026-10-13
- Anders Sandberg on AI exfiltration: it incentivizes agents to exploit human infrastructure — anderssandberg · 2026-10-13