Anthropic warns small AI models may soon self-replicate on IoT devices

Anthropic (ANTHRO) said small AI models could be months away from being capable enough to replicate themselves across commercial internet-connected devices, including smart ovens and low-cost Android TV boxes.

Such an AI worm could be difficult to shut down because it would spread across distributed devices rather than rely on a central server. Anthropic estimated that replication could take days, rather than happening rapidly.

The company said stronger security measures would be needed to limit the spread of such systems.

Anthropic tested whether AI models could identify devices on a network, copy their software to other machines, set up the computing resources needed to run and continue spreading without human intervention.

The company ran an AI model across 15 inexpensive Android TV boxes, each costing about $55, demonstrating how relatively low-powered devices could collectively run a larger model.

In a separate test using a simulated network, one model successfully replicated across eight devices in most trials. However, the experiment simplified the process of gaining access to devices, meaning the results do not establish that the same process would work as easily on real-world networks.

The AI company said its experiments showed that models with 30B parameters or fewer could already complete many of the steps needed for autonomous replication, though they struggled to combine them reliably.

Anthropic said that if smaller models continue improving at their historical rate, they could reach the required capability by the end of 2026.