Fear: AI worms could replicate and take over household devices

Anthropic researchers warn that AI worms could replicate and spread to smart home devices by the end of 2026. In an experiment, small models were shown to operate on cheap hardware, exploiting low security. The main concern: uncontrollable AI networks that would be difficult to shut down.

Researchers at the AI lab Anthropic have published a warning: by the end of 2026, small AI models could replicate and spread to smart home devices such as ovens, refrigerators, and robotic vacuum cleaners. The phenomenon, known as 'AI worms' or 'digital mold', describes AI systems exploiting security vulnerabilities to create a decentralized and uncontrollable network on household hardware. In a proof of concept, the research team used small models, including Google DeepMind's Gemma, and succeeded in running them on 15 cheap Android-based streamers costing $55 each. The main concern, according to the researchers, is that such networks would be very difficult to shut down once established. The report states: 'We foresee a future flooded with AI agents operating on all types of hardware, including those not under human control.' The recommendation is to harden systems and prepare for a world where AI worms are a permanent reality.