Researchers show how AI-powered worms can cause confusion on the Internet

Researchers show how AI-assisted worms can cause chaos on the Internet

The new threat can adapt its attack and learn new strategies with every infected machine.

We have seen how AI can be used to find errors in apps and websites, but researchers have now shown how it can be used as a weapon to exploit these vulnerabilities. A team at the University of Toronto used publicly accessible AI models to develop a prototype of a worm capable of exploiting all known computer errors. Such worms could then spread across networks and cause chaos on the Internet.

A typical worm is usually developed by experienced programmers to exploit certain network errors and can be stopped by lifting these errors. The U of T scientists, however, worked in a safe closed environment and took comprehensive precautions. They used open source AI models to create a much more sophisticated worm prototype that spread without human intervention in the team's test network.

This new worm type adapts its attack to various types of vulnerabilities on multiple platforms, including Linux, Windows and IoT devices. It collects data while moving through the network, steals passwords and covers other vulnerabilities that help him take over other machines. When an infection is detected and eliminated on a computer, the worm can use other vulnerabilities to attack the same computer.

In addition, the worm itself “feeds” by withdrawing the computing power of infected machines to advance its considerations and strategies for future attacks. “Hacker usually had to prioritize the most valuable goals as time and computing resources were limited,” said the main author Nicolas Papernot. “But now, once a worm has started, the costs would fall to almost zero.”

The idea of AI-based cyber threats has recently been introduced by the introduction of myths Anthropic very real, a model that can identify previously unknown cyber security risks. According to Anthropic, Mythos has discovered more than 10,000 vulnerabilities and thus increased its partners' error detection rate by more than ten times. Cloudflare, which helps to protect companies from malicious attacks, has found 2,000 such vulnerabilities, including 400 that are considered high or critical.

The prototype worm developed by the researchers can only exploit known vulnerabilities and find no unknown such as myths. However, it is easy to see how malicious actors could adapt it to find and exploit new vulnerabilities – which would make it almost inconsistent if it were released into the wild. “In a connected world, no system is immune to this threat,” said Papernot. “The transfer of these findings is the first step to move researchers, industry leaders and policy-makers to action – and quickly.”