Anthropic Warns Small AI Models Could Enable Self-Replicating IoT Malware

Anthropic is warning that advances in small AI models could soon enable self-replicating malware capable of infecting smart fridges, vacuum cleaners, and other Internet of Things devices. By next year, the company says, smaller AI models may autonomously self-replicate on IoT hardware, eventually enabling computer worms that spread without human interaction and can infect unfamiliar devices.

The company notes that such a replicator would be challenging to shut down because it lacks a centralized off switch, though it may spread slowly with a doubling time of days. Unlike non-AI malware, it could also figure out unfamiliar devices on its own. Anthropic cautions that its recent experiment reveals significant challenges in actually building such AI malware, tempering the findings.

In testing, Anthropic researchers ran small open-weight models on real or simulated low-cost Android TV boxes and other IoT hardware. Using quantization and workload sharding, the models achieved inference, with the Gemma4-26B-A4B model requiring just 15 boxes with 4GB of memory each. The Qwen3.5-9B model configured other LLMs to run on eight of nine hardware types, adapting configurations on the fly rather than needing device-specific code.

Read More at the original source →