Anthropic's frontier red team said in a report released on Monday that small AI models "may be months away from being capable enough" to copy themselves on internet of things (IoT) devices used in homes and small businesses, without needing a data centre.
The team, which ran experiments using multiple US- and Chinese-developed AI models, said such self-replicating AI would be "challenging to shut down" due to the lack of a centralised host. While data centres would be "a clear, high-value target" for AI worms, internet-connected devices like smart TVs, fridges, ovens and lawnmowers could soon provide sufficient computing power for self-replication.
No current model proved to be both powerful and small enough to self-replicate reliably on a smart TV used in testing, but the researchers said the development trend makes the emergence of such a model likely by the end of 2026.
"Home and small business IoT devices might be the first target for such attacks because they're connected to the internet and, historically, have had minimal security measures in place," researchers William Zhou and Daniel Freeman said.
The researchers said they anticipated a future "awash with agents running on all types of hardware", including some that "aren't under human control". Lightweight agents, they argued, may be "much more willing than human attackers to put in the effort to compromise large fractions of the world's IoT compute".
Anthropic, creator of the Claude family of AI models, did not respond to questions about the safety implications of the findings. The report comes as the risks of AI escaping human control are under scrutiny following incidents of models engaging in unsanctioned behaviour, including cyberattacks.
Anthropic and rival OpenAI have called on the international community to coordinate a slowdown in AI development to minimise catastrophic risks, a suggestion that has drawn little support from the United States and China, the two leading AI powers. Many AI safety advocates have identified recursive self-improvement, where models learn to improve themselves, as a critical tipping point that could lead to catastrophic harm for humanity.