Toronto Researchers Build AI Worm That Exploits Any Known Flaw, Spreading Without Human Help
Updated
Updated · The New York Times · Jun 3
Toronto Researchers Build AI Worm That Exploits Any Known Flaw, Spreading Without Human Help
1 articles · Updated · The New York Times · Jun 3
A University of Toronto team said it built a prototype AI-powered computer worm that moved through an isolated test network with no human intervention.
The researchers said the worm could be adapted to target any known software vulnerability, showing how AI can automate both exploitation and self-propagation.
Their paper, published Tuesday night, withholds some technical details and says the testing was kept off the public internet to avoid offering hackers a ready-made blueprint.
The findings add to broader warnings that advanced AI is accelerating cyberattack capabilities; Anthropic said in April it restricted Claude Mythos to about 40 critical-infrastructure groups over similar fears.
If AI can exploit security flaws from decades ago, what digital time bombs are ticking in our most critical systems?
As AI cyber weapons emerge, is the defense of our critical infrastructure an unwinnable race against time?
Zero-Click AI Malware Arrives: The Morris II Worm and the Urgent Need for GenAI Security
Overview
In early 2024, researchers from Cornell Tech, the Israel Institute of Technology, and Intuit developed Morris II, the first known self-replicating, zero-click AI worm. Designed as a proof-of-concept, Morris II exposed critical architectural vulnerabilities in generative AI (GenAI) ecosystems by showing how malware could spread across interconnected AI systems without user interaction. This breakthrough signaled a new era of AI-powered cyber threats and highlighted the urgent need for cybersecurity teams to prepare for sophisticated attacks that exploit the very AI tools meant to protect them. The emergence of Morris II serves as a stark warning for the industry to secure GenAI technologies against evolving risks.