Researchers Develop New AI-Powered Worm That Might Be Unstoppable

Key Points:

• Researchers from the University of Toronto, University of Cambridge, and others have demonstrated an AI-powered malware "worm" that autonomously spreads across devices by tailoring attack strategies to each target using publicly available AI models.
• Unlike traditional worms like WannaCry, which exploited a single vulnerability, this AI worm dynamically detects unique security flaws on each device, making it more adaptable and harder to stop.
• The worm exploits the increasing computational power of modern devices built to support large language models (LLMs), using infected devices' resources to propagate further across networks.
• Although slower than traditional worms—taking about five days to infect half the devices in a test network—its speed is expected to increase as AI models and device efficiency improve.
• The researchers published their findings cautiously, omitting key details to avoid misuse, and aim to alert the cybersecurity community to this emerging threat amid ongoing efforts by organizations like Anthropic and OpenAI to develop AI tools for cybersecurity defense.