Anthropic red team built AI worm that spread across smart TVs
The offensive research team split the program across 15 internet-connected TVs, each lending a little computing power, The Guardian reports.
Anthropic's offensive research team built a malicious AI program that spreads autonomously through internet-connected smart TVs, with the program splitting itself across 15 televisions and borrowing a small amount of computing power from each device, The Guardian reported on October 18, 2026.
The researchers said the same approach could work on other ordinary devices. They argued that smart devices receive security patches less often than phones or laptops, which would make AI worms harder to stop.
The team also raised the hypothesis that an AI worm could investigate newly disclosed vulnerabilities on its own and move on to other hardware. They cautioned that AI worms would need to perform more tasks on lower-powered devices than is possible today, and that the technology has not reached that level — though they said it could get there within the year at the current pace of development.
Security experts divided on the risk
Ariel Herbert-Voss, who leads the AI security startup Lunasec, said AI models differ from conventional viruses because they learn as they move. Joshua Saxe, who led AI security development at Meta, said he worries the security industry does not take the real risk seriously enough.
Jonas Guan, a doctoral student at the University of Toronto, said the Anthropic worm moves very slowly and its behaviour can be predicted and detected, allowing it to be blocked at multiple stages. Juan Andres Guerrero-Saade, vice president of research at SentinelOne, called it a stunt hack that promotes absurd doom scenarios about AI models having their own will. Experts said basic practices such as strong passwords and two-factor authentication can stop most worms.
Anthropic did not respond to the Guardian's interview request. Worms such as NotPetya and WannaCry caused enormous damage worldwide before this research.
Quick answers
What did Anthropic's red team build?
A malicious AI program that spreads autonomously across internet-connected smart TVs; the researchers split it across 15 televisions, drawing a little computing power from each.
Can the worm be stopped?
Jonas Guan of the University of Toronto said it moves slowly and its behaviour can be predicted and detected, so it can be blocked at multiple stages. Experts said strong passwords and two-factor authentication stop most worms.
Did Anthropic comment on the research?
Anthropic did not respond to the Guardian's interview request.