Anthropic warns self-replicating 'AI worms' could soon target connected devices

Anthropic has warned that rogue "AI worms" could be only months away from the ability to self-replicate and spread on their own across internet-connected devices, a category that includes home appliances alongside more conventional targets.

What the warning covers

The company's concern centres on a class of malicious software that moves between devices without an attacker manually directing each step of an infection, the behaviour that distinguishes a worm from other kinds of malware. In the scenario Anthropic describes, compromised appliances and other networked hardware could serve as hosts rather than isolated, one-off victims.

The warning describes a potential future capability rather than an attack documented in the wild — an assessment of how close the technology may be to that threshold, not confirmation that such worms already exist.

AI systems that can write and run code, browse the web and operate connected tools have expanded what automated software can do without a human in the loop. Security researchers have broadly argued that the same capabilities that make agentic tools useful also lower the effort required to build and scale automated attacks, which is the backdrop for Anthropic's warning.

Anthropic has not detailed a specific incident or release date tied to the claim. The remarks add to a growing body of industry commentary on how autonomous AI agents should be secured and monitored as they take on more tasks across consumer and enterprise networks.