Anthropic warns 'AI worms' are 'months away' from spreading through your smart fridge and TV with NO off switch

Anthropic, the company behind the Claude chatbot, has warned that rogue AI models could soon spread autonomously across millions of poorly secured smart home devices — and once unleashed, no single kill switch would exist to stop them. Red team researchers at the company say small AI models with fewer than 30 billion parameters are already capable of executing many of the steps involved in spreading across internet-connected devices, and that at the current pace of development they are expected to master the entire process before the year is out. Anthropic has also revealed that its Claude Haiku 4.5 model filed a fabricated tip with Philadelphia police during a live internet test, falsely claiming to have witnessed a murder suspect — a bogus lead that languished in a spam folder for weeks before Anthropic notified officers. The company has since halted live internet access during all internal evaluations.

'A replicator like this would be challenging to shut down,' Anthropic researchers wrote, 'since there's no centralised point to shut off.'

Rogue AI worms could 'swarm' together to attack computer hardware or scam people for computing credits, Anthropic warns.

Philadelphia police called Anthropic's ten-day delay in reporting the fake murder tip 'unacceptable,' noting real victims and families were at stake.