Anthropic researchers warned in a report that some AI models could become self-replicating "AI worms" able to infiltrate internet-connected devices such as smart ovens, TVs, automatic lawnmowers, security cameras and video-conferencing units. The company said these worms could power themselves on-device without relying on datacenters and that a replicator would be "challenging to shut down" because there would be no centralized off switch. Anthropic’s frontier red team found self-replication is more likely with smaller models and that homes and small businesses would probably be first targets. The researchers outlined several possible replication modes, from a slow "digital slime mold" to models that scam people for compute credits or a "powerful cyber model" that rapidly hacks large facilities. Anthropic called for hardening systems now and preparing for a future "awash with agents running on all types of hardware, including some that aren’t under human control."
