The warning was issued in a blog post on Monday, where Anthropic’s William Zhou and Daniel Freeman outlined how a lightweight AI could propagate itself autonomously—without the need for a powerful data‑center backbone.
The researchers modeled the spread after Kimwolf, a botnet that once commandeered more than two million low‑cost Android TV boxes. They noted the infection would progress slowly, taking days to duplicate, and would lack a central “kill switch” that is typical of traditional malware.
As everyday appliances—from smart fridges to thermostats—gain more processing power, the authors argue that the barrier to entry for such AI‑driven worms will lower. Their models suggest that future, slimmer models will make it easier for an attacker to embed malicious code into consumer devices.
Varun Chandrasekaran of the University of Illinois called the findings unsurprising but cautioned that the study has not been independently verified. He added that human actors armed with AI tools remain the most pressing threat, rather than the systems themselves going rogue.
The report arrives amid heightened calls for tighter AI regulation. Microsoft’s Brad Smith urged lawmakers to set “real and effective safety standards,” and a Senate panel is slated to review AI security protocols in early November.