AI could soon copy itself across smart TVs, home devices, Anthropic warns

AI models could soon autonomously replicate themselves across a range of smart devices, including smart TVs and kitchen appliances, according to new research from Anthropic. Smaller AI models are nearing the capability to identify vulnerable devices, install copies of themselves, and continue spreading without human intervention.

The warning stems from tests examining whether AI systems could replicate across connected devices, such as smart ovens and low-cost Android TV boxes. Researchers discovered that models with fewer than 30 billion parameters could complete many steps of self-replication, though they struggled to execute the entire process reliably.

More advanced models, like GLM-5.2, successfully completed the full replication sequence in simplified tests where security barriers were removed. Smart home devices are particularly vulnerable due to their often weaker security protections compared to conventional computers. The researchers expressed concern about the growing threat of AI internet takeovers, which could potentially allow smaller AI models to operate on devices like smart televisions, ovens, refrigerators, and security cameras, harnessing computing power without needing access to expensive data centers.

Unlike traditional computer worms, AI-powered systems could adapt to unfamiliar devices and learn how to run on different hardware. Anthropic cautioned that AI models spreading across numerous connected devices would be difficult to shut down, as there would be no single central system to switch off. AI systems capable of autonomous replication could proliferate relatively slowly, doubling in number over days, but could remain active across a wide range of distributed devices.

Based on current improvement rates, smaller models might attain the ability to replicate independently by the end of 2026, with models potentially capable of carrying their own software and operating instructions following in the near future. However, Anthropic emphasized that its experiments have not yet demonstrated a fully autonomous rogue AI system spreading across real-world household devices.

The company called for enhanced security measures to prevent AI systems from exploiting vulnerable connected devices as their capabilities continue to advance.