A prototype worm developed by researchers at the University of Toronto demonstrates the potential for AI to exploit computer vulnerabilities in a novel manner. This worm can adapt its attack strategies as it spreads through networks, targeting various platforms such as Linux, Windows, and IoT devices. Unlike traditional worms that rely on pre-programmed attacks, this AI-driven model collects data about the network and its weaknesses, allowing it to siphon passwords and gain access to additional systems.
The worm operates autonomously, requiring no human intervention, and can continue to exploit new flaws even after initial defenses are patched. As it infects machines, it utilizes their processing power to enhance its reasoning capabilities for future assaults. According to lead author Nicolas Papernot, this advancement significantly reduces the cost of such attacks, shifting the dynamic of cybersecurity threats.
This development comes in the wake of Anthropic's launch of Mythos, an AI model that has identified over 10,000 cybersecurity flaws. Although the prototype worm currently can only exploit known vulnerabilities, its design raises concerns about the potential for malicious use in the future.