Anthropic Research Tests Feasibility of Self-Replicating AI

Research highlighted by Better Stack examined whether AI agents could autonomously replicate across networked devices. Experiments on cheap Android TV boxes showed potential but exposed major barriers: limited compute, model size, reliance on external servers and slow replication.

• A prototype worm distributed larger models across multiple devices
• Barriers include weak hardware, model size and server dependency
• ISPs can easily detect and block unusual network traffic
• Risk stems from open debug ports and weak passwords in IoT