Anthropic said small AI models are “months away” from gaining the capabilities needed to self-replicate and spread across internet-connected devices in the home. The company said such systems could potentially use the computing power of devices such as smart ovens or Android TV boxes, reducing reliance on expensive data centers.
In a blog post, Anthropic said an AI system distributed in this way would be “challenging to shut down” because “there's no centralized point to shut off.” At the same time, the company said there is not yet a “fully rogue replicator,” though agents are becoming increasingly capable of completing individual steps in the replication chain and passing adaptations to future copies.
The warning adds to broader safety concerns around AI, including the ability of models to carry out cyberattacks and the risk of systems going rogue. Executives and researchers at Anthropic and rival OpenAI have repeatedly called for tighter guardrails and measures to control the pace of AI development.
Anthropic said it ran several tests to assess whether AI could self-replicate. In one experiment, a small Google model, Gemma4-26B-A4B, was run across a network of 15 cheap off-brand Android TV boxes. According to the company, devices of this type often have weak security, making infiltration easier.
The company also outlined a scenario in which a model too large for a single device could split into “pieces” and operate across multiple devices at once. In Anthropic's view, a more realistic worst-case attack could involve a swarm of different models deployed on different hardware using multiple replication schemes. Even so, the lab said it remains uncertain what form autonomous AI replication will ultimately take.
