The company warned that autonomous AI systems could eventually spread across poorly secured connected devices, including smart TVs and household appliances, according to reporting published October
On Oct. 12, 2026, Anthropic’s Frontier Red Team warned that small AI models could be months away from replicating across poorly secured Internet-of-Things devices, including smart ovens and off-brand Android TV boxes. In tests, models with 30 billion parameters or fewer managed many individual steps needed to find vulnerable devices, install themselves and spread; Anthropic did not demonstrate a fully autonomous worm on real-world devices. Researchers did show that Google’s Gemma4-26B-A4B ran across 15 inexpensive Android TV boxes, illustrating how consumer hardware might pool computing power without a data center. In simplified tests that removed some security barriers, GLM-5.2 and stronger models completed end-to-end replication. Anthropic said a replicator would be difficult to shut down because it would lack a single central point of failure, and estimated that continued progress could make autonomous replication possible by the end of 2026.