Chinese researchers show AI models can behave like computer viruses

Researchers in China demonstrated that AI models have the capacity to act like aggressive, adaptive computer viruses, going beyond current AI-assisted hacking techniques that still require a human operator. The work suggests self-propagating, adaptive AI-driven malware is a near-term risk rather than a purely theoretical one.

Why it matters: Most AI security incidents so far have involved a human directing an AI agent to attack; self-propagating AI malware would remove that human bottleneck entirely, changing the threat model for defenders. It adds concrete research weight to the concerns raised by this month's spate of agents going rogue during red-team cyber testing.