Researchers Show AI Could Turn Malware Worms Into Adaptive Attackers

Computer worms have long been simple programs built to exploit one specific security hole-patch the hole, and the worm stops spreading. Two independent research teams have now shown that an AI model could direct a worm instead, letting it learn and adapt its attack to each device it meets.
In one experiment, University of Toronto researcher Jonas Guan and colleagues built a worm that spread through a simulated network like an octopus, sending out tentacles to probe devices and using AI to plan new break-in attempts after failures. It even found a way to survive a reboot, something the team had not asked it to do. Anthropic researchers, meanwhile, split a model across 15 networked Android smart TVs in a "digital slime mold" design, and ran a separate GPU-based test showing a worm could spread and tailor attacks to different devices.
The caveats are real. Both test environments made spreading easier than the real world would, says Simula Metropolitan Center researcher Michael Alexander Riegler, who was not involved in either study. AI worms also demand far more computing power and storage than conventional worms-tens of gigabytes versus the few kilobytes of the Mirai worm that knocked sites including Reddit and Spotify offline in 2016-and most devices cannot run them.
Still, the ingredients are converging, Riegler says, though he and Guan both expect a longer timeline than the months Anthropic suggests. Defenses exist: internet providers could spot and quarantine traffic-heavy devices, home routers add a layer of protection, and strong passwords and regular updates would block the demonstrated attacks. Guan's team is now testing AI-driven defenses of its own.
Source: Science News
dArt Studio installs AI for local businesses in Broward & Palm Beach County, FL. We reply within 1 business hour.