University of Toronto researchers warn AI could power adaptive cyber worm

Researchers at the University of Toronto said they had shown how publicly accessible artificial intelligence models could power a self-adapting cyber worm, a finding that raises the stakes for already strained defenses across connected systems. The university said the threat could be built with free AI models, could seize control of an entire network, and could hijack computing power for hackers at virtually no cost.

The work, released June 2, 2026, was conducted in a secure digital lab and was presented as a warning shot for the cybersecurity field. U of T said the worm would not rely on a single fixed attack path. Instead, it could adjust its strategy as it moved from one device to the next, making it harder to stop with static filters, patch cycles, or signature-based tools. Every online device was described as a potential target, a reminder that the attack surface stretches far beyond corporate servers to home machines, cloud systems, and the devices that quietly hold together hospitals, schools, and public agencies.

Nicolas Papernot, an associate professor of electrical and computer engineering, computer science, and law at the University of Toronto and a faculty member at the Vector Institute, was named as the lead researcher in the university’s coverage. Papernot’s research sits at the intersection of computer security, privacy, and machine learning, a mix that reflects how quickly the defensive and offensive sides of AI are converging. The team included Gururaj Saileshwar, Chris (Shaopeng) Lin, Joyce Qu, and other researchers working inside U of T’s broader AI ecosystem.

The warning echoes the Morris Worm, which the FBI says was unleashed from the Massachusetts Institute of Technology on November 2, 1988, at around 8:30 p.m. and spread rapidly across the internet, grinding computers to a halt. That episode became one of the early reminders that a single self-replicating program can outpace human response. The University of Toronto work suggests AI could make that problem more adaptive, more scalable, and cheaper to launch.

The gap, for now, is preparedness. U of T said current cyber defenses are not ready for this threat, even as attackers could use free models to automate reconnaissance, tailor exploits, and spread laterally through networks. The university has been trying to prepare for that broader shift: it established an AI Task Force in spring 2024 to develop a comprehensive vision and strategy for generative AI. The new research shows how quickly that strategy has to move, because the next worm may not just spread. It may learn.