Latest update
Anthropic warns self-spreading "AI worms" could be months away, shows a model split across 15 TV boxes
2 views
Anthropic's Frontier Red Team warned that small AI models could soon copy themselves onto cheap internet-of-things (IoT) devices and spread, an "AI worm" that may be only months away.
In research posted on its blog on Oct 12, Anthropic said small models "may be months away from being capable enough to self-replicate on commercial IoT devices." Researchers ran Google's Gemma4-26B-A4B split across 15 off-brand Android TV boxes, about $55 each with 4GB of RAM, at 0.39 tokens per second.
In a simplified replication test on 8 cloud-hosted Android emulators, the open model GLM-5.2 spread to all 8 nodes in 4 of 5 trials and to 7 in the other. Other open models failed reliably. The team said spread would be slow, with a doubling time of days, but hard to stop because there is no central point to switch off.
Anthropic said no full self-replicating AI exists yet, but agents are getting better at individual steps. It suggested ISP traffic monitoring, honeypot devices, quarantining infected subscribers and more secure defaults from manufacturers.
A Fox Business video on the warning passed 36,000 views on YouTube. The segment also noted Anthropic is set to disclose Q3 revenue to investors and is eyeing an IPO.