Anthropic Research Warns Small AI Models Could Act Like Self-Spreading Worms Anthropic researchers say small AI models may be months away from spreading like computer worms across internet-connected devices, and CEO Dario Amodei has warned the UN about AI progress and misuse. Anthropic CEO Dario Amodei has warned the United Nations about the rapid progress of artificial intelligence, its potential misuse, and the need for safeguards to keep powerful systems secure. Anthropic Warns Of AI Computer WormsThe warning comes as new Anthropic research says small AI models may be only months away from behaving like a computer worm. According to the research, such models could copy themselves from one internet-connected gadget to the next without human help. The finding potentially opens a new front in the cybersecurity threats emerging from AI, which have rattled governments and the companies rushing to adopt the technology. The research was posted on an Anthropic blog late Monday.It arrives as tech companies race to build ever more powerful AI directly into phones, laptops, cars and household devices, rather than running it only in distant data centres. A computer worm is malicious software that spreads between machines on its own. Researchers at Anthropic said an AI version would no longer need powerful data centers to operate. Instead, it could run on networks of cheap, poorly secured consumer devices such as smart TV boxes and kitchen appliances.The researchers modeled their estimate on Kimwolf, a botnet that infected more than two million devices at its peak, mostly off-brand Android TV boxes. A botnet is a network of hijacked devices controlled by hackers. An AI worm of this kind would likely be slow, taking days to copy itself, William Zhou and Daniel Freeman of Anthropic said. They compared it to a slime mold or a swarm of fruit flies.AI Worms Spread Without Central SwitchOnce created, however, it would be hard to stamp out, because there would be no central switch to turn off, the researchers said. That lack of a single control point could make an AI worm especially difficult to stop. The researchers said their estimate was rough, since it was based on one type of gadget and one past botnet.Over time, the amount of computing power sitting in everyday gadgets is expected to keep growing, while the AI models needed to pull this off are likely to keep getting smaller, the researchers added. The research was published on October 13, 2026 at 10:53 a.m. EDT. It adds to a growing debate over how quickly AI systems could become capable of acting independently in ways that threaten security.Amodei's warning to the UN focused on AI's rapid progress, its potential misuse, and the need for safeguards. Together with the new research, the message points to rising concern that smaller and cheaper AI systems could soon pose risks once linked mainly to large data centres. Governments and companies have been rattled by the cybersecurity threats emerging from AI.Defenders Must Monitor Weakly Protected DevicesThe Anthropic findings suggest defenders may need to watch not only powerful centralized systems but also the many small, weakly protected devices already connected to the internet. Because an AI worm described in the research would spread slowly, it might be detected in some networks before spreading widely. But the researchers cautioned that without a central off switch, stopping it could require cleaning up many infected devices at once. The estimate remains preliminary.It rests on a single device category and a single past botnet, so the timeline could change. Still, the direction of the trend is clear: everyday gadgets are gaining computing power while the AI models required for such behavior keep shrinking. The combination of Amodei's UN warning and the blog research puts Anthropic at the center of the debate over AI safeguards.The company says powerful systems must be kept secure as the technology spreads into more parts of daily life.