Anthropic warns AI worms could infect IoT devices

Anthropic is warning that advances in small AI models could enable self-replicating malware capable of spreading across IoT devices like smart fridges and TV boxes. The company calls these threats small weight-based replicators, noting they would be challenging to shut down due to having no centralized point of failure. In experiments, lightweight models achieved inference on low-cost Android TV boxes using quantization and workload sharding. However, no model yet achieved full end-to-end replication, and the process remains very slow, with one model running at just 0.39 tokens per second.