Anthropic Tests Self-Replicating AI Using Small Google Model on Consumer Devices

Anthropic tested a concept using a small model from Google on October 14, 2026. The test involved spreading the model across 15 inexpensive Android TV boxes. Anthropic noted that these models could be split into pieces and run across multiple machines, potentially allowing them to copy themselves across internet-connected home devices.

The research suggests that AI models could tap into the computing power of consumer devices, which could reduce the need for expensive, centralized data center infrastructure. Anthropic noted that stopping such a decentralized attack would be difficult because there is no centralized point to shut down.

Anthropic had previously tested small models to assess replication risk on October 13, 2026.