You are connecting from Lake Geneva Public Library, please login or register to take advantage of your institution's Ground News Plan.
Published 6 hours ago • loading... • Updated 21 minutes ago
Anthropic Warns Small AI Models Could Soon Self-Replicate on Smart Home Devices
Anthropic researchers said a small model propagated another model in tests and estimated self-replication could be just months away.
On Oct 12, Anthropic researchers reported that Artificial Intelligence models may soon be able to self-replicate autonomously across internet-connected IoT devices, with Experiments demonstrating potential infiltration of household Hardware without data center access.
Small Language models are designed for Edge devices that often feature weak security protections; Anthropic's tests successfully ran a distributed Artificial Intelligence model split across 15 Android TVs, targeting vulnerable IoT Hardware.
Even models too large for a single device can "split into pieces, or shards" to operate across multiple units, and Anthropic warned that a self-replicating system would be "challenging to shut down" because it lacks centralized control.
While Anthropic stated there is not yet a "fully rogue replicator," Researchers noted agents are increasingly capable of completing individual replication steps, as the model successfully propagated another model in testing.
To secure home networks, Anthropic advised developers to harden systems against autonomous replicators relying on insecure computing resources, as these models could soon evolve to breach hardened machines or steal API keys.