AI is advancing rapidly, with researchers at Anthropic’s Frontier Red Team cautioning that users might start seeing replicated versions of small models on IoT devices like smart TVs "within months."
On Monday, the team said it is already "seeing signs of life for AI worms" with limited capabilities, although there is room for improvement. They can scan their network, execute an exploit, and copy model weights to a target machine across a GPU cluster.
"We don’t yet have a full rogue replicator, but agents are becoming increasingly capable of completing the individual steps in the replication chain (and passing on their adaptations to future copies of themselves)," the team wrote.
Anthropic said models with 30 billion parameters or fewer can already handle many of the steps. However, GLM-5.2 completed end-to-end replication in four of five trials across a firewalled virtual private cloud of eight nodes.
Separately, the team also tested a 26-billion-parameter model across 15 off-brand Android TV boxes, with the researchers observing that weak security on such IoT hardware makes it an appealing target for anything that copies itself by its model weights.
"While it’s true that IoT devices have been vulnerable for a long time, lightweight agents may be much more willing than human attackers to put in the effort to compromise large fractions of the world’s IoT compute," the team said.
Drawing on the Epoch Capabilities Index, the analysis projects that small models will be able to replicate themselves by the end of 2026.
An AI-driven IoT worm would not behave like fast-moving malware. Instead, it would look like a digital slime mold creeping across poorly protected edge devices, Anthropic said.