Anthropic warns small AI models could self-replicate across IoT devices

Anthropic says small AI models may be months away from being capable of self-replicating across commercial internet-of-things (IoT) devices, such as smart ovens and off-brand Android TV boxes. The company’s experiments showed that models could identify unfamiliar devices, connect them to a network and prepare them to run an AI model.

Researchers distributed Google DeepMind’s Gemma4-26B-A4B across 15 off-brand Android TV boxes. They did not demonstrate a fully autonomous, self-replicating AI worm on real IoT hardware. The tests used a simplified setup in which target devices exposed unauthenticated ADB access.

In tests across nine hardware types, Qwen3.5-9B installed and configured software to run Llama3.1-8B on eight. GLM-5.2 reached all eight nodes in a controlled emulated network in four of five trials, while other models did not reliably complete the process. Qwen3.5-9B’s success rate rose from 45% in early attempts to 84% in later generations when data from previous attempts was retained. Assistance from Opus 4.8 produced no measurable improvement.

Based on these results and historical data, Anthropic estimates that models with 30 billion parameters or fewer could reach the required capability by the end of 2026. On the studied hardware, inference took 28 to 51 hours for Qwen3.5-9B and 71 to 150 hours for Gemma4-26B-A4B. Anthropic said the findings do not indicate an imminent threat and did not demonstrate full end-to-end worm creation.