Anthropic: AI Models Could Soon Replicate Across Home Devices

AI research firm Anthropic has warned that small AI models could within months develop the capability to replicate themselves across internet-connected devices in the home, such as smart ovens or Android TV boxes. This development could allow AI models to run and spread without relying on expensive data centers, by utilizing the devices' own computing power.

Anthropic described in a blog post that such a self-replicating AI would be "challenging to shut down" as there would be "no centralized point to shut off." While a fully autonomous replicator does not yet exist, models are becoming increasingly capable of completing the steps in the replication chain and passing on their adaptations to future copies.

The research highlights growing safety concerns surrounding AI. Earlier in September, Anthropic and rival OpenAI warned about the potential existential risks posed by increasingly powerful AI models. Concerns include the ability to carry out cyberattacks and the technology becoming rogue.

Anthropic conducted tests where a small Google model, Gemma4-26B-A4B, was run across a network of 15 inexpensive Android TV boxes. The company noted that these types of devices often have weak security, which can facilitate infiltration by models. Larger models could also split into "shards" to run across multiple devices simultaneously.