New Delhi: In what can be a nightmare for cybersecurity experts, researchers in Canada have developed an AI-powered computer worm that can adapt to attack any online device.
In a study, whose findings were published on the preprint server arXiv.org on 2 June, researchers from the University of Toronto demonstrated how hackers can use cheap and free-to-use AI models to power computer worms that change their strategy to infect entire computer networks.
This comes as Anthropic’s new AI model Mythos raises cybersecurity concerns globally, with its ability to find security flaws in computer systems and software, including those that have eluded detection for decades.
A traditional computer worm can replicate itself and move from one computer to another without any user intervention. It infects systems by identifying a vulnerability and usually spreads through mail attachments. But with AI, this worm can now learn and adapt to spot different types of vulnerabilities, making it tricky to curb its spread.
“Traditional worms can be stopped by patching the specific vulnerability they exploit. Our adaptive worm cannot be stopped this way: it uses a recursive reasoning loop to detect and exploit diverse vulnerabilities as it propagates,” researchers said in a statement.
Also read: Why Anthropic and OpenAI’s latest warnings sent shockwaves
How it works
The study was conducted in a secure digital lab, and national science, security, and defence bodies were consulted before the findings were published. According to Nicolas Papernot, an Associate Professor at the University of Toronto and an author of the research, it was important to develop the worm “in a controlled, academic setting before bad actors figured it out for themselves”.
To assess the capability of the AI-driven worm, researchers ran it through a simulated 33-machine corporate network 15 times. And in one week, the malware had penetrated about 75% of the machines without any human involvement.
Once inside a computer network, the worm uses the processing power of the system for reasoning and then identifying its next target — any device connected to the Internet. This also brings down the cost of a cyberattack.
“Hackers have typically had to prioritize the most high-value targets because time and computing resources were limited But now, once a worm is launched, the cost would drop to nearly zero,” said Papernot.
He warned that no computer system is immune to the threat and the study is the “first step in galvanizing researchers, industry leaders and policymakers to take action – and quickly.”