Worms differ from ordinary viruses because they need not trick a person to spread, and earlier ones have caused some of the worst damage in internet history: the Morris worm of 1988 slowed many infected machines to a halt, Stuxnet sabotaged Iranian nuclear centrifuges, and NotPetya and WannaCry caused tens of billions of dollars in damage worldwide. The Anthropic team suggested an AI worm could exploit unpatched vulnerabilities in the same way but be harder to stop, because smart devices receive patches less often than phones or laptops, and could keep adapting to new hardware by researching newly disclosed flaws.

Not everyone is convinced the threat is imminent. Jonas Guan, a Toronto PhD student who co-authored an earlier paper in which a worm infected a full-size computer and spread to similar machines, said the Anthropic version moves very slowly and that worms have predictable, detectable actions that can be blocked at multiple layers. SentinelOne research vice-president Juan Andres Guerrero-Saade called it, in his view, "nifty stunt hacking" rather than something truly feasible or impactful. Experts added that basic hygiene, including strong passwords and two-factor authentication, would stop most worms.

Others see the demonstration as a warning worth heeding. Chris Inglis, a former National Cyber Director, said the question of whether AI worms are plausible "has been settled by several breakout events over the past six months", and that the better question is how they will be employed and countered. Joshua Saxe, who led Meta's AI security effort, said he is concerned the cybersecurity community is not taking these real risks as seriously as it should.