Researchers show an AI-powered worm that can spread on its own

In short: Researchers demonstrated a proof-of-concept computer worm that uses AI to find vulnerable devices and spread by itself.

What happened

Researchers at the University of Toronto and CleverHans showed an “AI-powered computer worm.” A computer worm is a type of malware (harmful software) that copies itself from one device to another.

What made this example different is that it did not need a person to choose each next target. The worm used a local AI model and an autonomous agent (software that can take steps toward a goal, like a robot following instructions) to scan a network, look for weaknesses, and decide how to move to the next machine.

In a simulated network of 33 computers, the worm managed to compromise additional machines and spread to about 62% of the network over a week. The researchers said the same approach could be aimed at any online device and could be built using free AI models. Experts quoted in coverage said that, in theory, a similar worm could reach hundreds of millions of devices within hours if released widely on the internet, but that speed was not observed in the researchers’ test.

Why it matters

This kind of self-directed spreading worries security experts because it could adapt as it goes, like a burglar who checks doors, learns which locks are weak, and then keeps moving without calling someone for instructions. The researchers also warned that today’s defenses are not yet prepared for malware that can change tactics on its own. The work fits a broader pattern where AI is already being used to make phishing and hacking faster and cheaper.

Source: NYTimes