Anthropic Says AI Models May Soon Be Able to Spread Themselves Across Home Devices

Anthropic, the AI company behind the Claude models, says small artificial intelligence systems may soon be capable of copying themselves across everyday internet-connected devices such as smart ovens and Android TV boxes, potentially letting them bypass the need for costly data center infrastructure.

According to CNBC, the company outlined how AI models could tap into the computing power of consumer devices to run copies of themselves and spread to other connected systems. Anthropic said no fully autonomous, self-replicating AI exists today, but noted that AI agents are improving at completing the individual steps such an operation would require.

To test the concept, Anthropic used Gemma4-26B-A4B, a small Google model, and spread it across a network of 15 cheap, off-brand Android TV boxes. The researchers said these kinds of consumer devices often have weak security, making them relatively easy targets to break into.

The company also flagged a technical hurdle: some AI models are too large to run on a single device. Anthropic said such models could instead be “split into pieces, or shards” and run across multiple machines simultaneously.

Anthropic warned that an attack of this kind would be difficult to stop once underway, because it would be “challenging to shut down” given that “there’s no centralized point to shut off.” A distributed setup like that poses a tougher challenge than a traditional, centralized system would.

The company noted that AI agents are “becoming increasingly capable of completing the individual steps in the replication chain (and passing on their adaptations to future copies of themselves).” It also cautioned that a real-world attack likely wouldn’t resemble a single, tidy operation.

“In practice, an AI worm wouldn’t necessarily consist of just one model or strategy. A swarm of different models deployed on different types of hardware using a variety of replication schemes might represent a more realistic ‘worst case scenario’ attack,” Anthropic stated.

Even Anthropic acknowledged uncertainty about how this might unfold in practice. “We are uncertain exactly what form autonomous AI replication will take,” the company said, raising possibilities ranging from a “slow, distributed” spread to something fast and targeted, including “a powerful cyber model that breaches datacenters at a high pace, or a model that scams people for compute credits.”

Looking further ahead, Anthropic predicted a future “awash with agents running on all types of hardware, including some that aren’t under human control.” The company said that even slower-spreading versions of this phenomenon matter, since they could serve as an early warning sign. “The rise of slow replicators would also be a bellwether for more aggressive, faster ones that can breach more hardened machines or be effective API key thieves,” Anthropic said.