AI Worm: Free LLM Powers Adaptive Network Attacks

A new AI worm can compromise nearly three-quarters of a simulated enterprise network in just seven days. This self-propagating worm uses free, publicly available large language models and requires no specialized skills or expensive computing. What's interesting is this worm doesn't rely on a fixed list of attacks. Instead, it analyzes each target, identifies vulnerabilities, and creates a customized attack. This allows it to adapt to new threats. Researchers at the University of Toronto demonstrated this capability. In a simulation, the worm found an average of 31.3 vulnerabilities per run and exploited 73.8% of the network. It even exploited vulnerabilities discovered after its own training. The bottom line is that the cybersecurity industry is now responding to threats from autonomous AI systems already in motion. This means new defenses are urgently needed.