Artificial intelligence experts are cautioning the general public to utilize robust passwords and promptly update software on their devices to counteract the emergence of “AI-driven computer worms,” a novel form of cyber threat that can execute customized attacks on devices, draining processing power and data as they seek out new targets.
In June, a team from the University of Toronto, headed by Canadian Institute for Advanced Research AI chair Nicolas Papernot, revealed that publicly accessible AI models have the potential to fuel a worm that can adjust its attack tactics on the go while spreading across internet-connected devices such as laptops, printers, and cameras.
The study, carried out in collaboration with the Vector Institute, was shared with various national science, security, and defense bodies before being made public. Papernot, an associate professor at U of T specializing in computer engineering and computer science, now advises the public to heed software update notifications promptly and change passwords regularly.
“We must no longer be lax in our cybersecurity practices,” Papernot emphasized during a panel discussion at U of T. “We must avoid password reuse, implement multi-factor authentication, ensure our devices are up to date, and organizations must adapt their processes to swiftly deploy software patches.”
Unlike traditional computer viruses, worms can spread automatically from one device to another without human intervention. The worm developed by the U of T researchers in a controlled setting gathers intelligence as it traverses devices, exploiting each breach to uncover passwords and vulnerabilities to target other machines.
In an unregulated environment, such a worm could gain internet access and assimilate information from alerts about newly identified vulnerabilities, surpassing the speed at which software updates are released to counteract them. The researchers highlighted that while some vulnerabilities can be mitigated through software patches, others, such as weak passwords and inadequate IT configurations, require human intervention and cannot be resolved solely through patch deployment.
Papernot cautioned that the AI-driven worms pose a substantial cybersecurity risk as they can devise customized attack strategies for individual victim devices, unlike traditional attacks that rely on a fixed script. This adaptability makes it challenging to halt the spread of such worms by addressing singular or limited vulnerabilities.
Papernot’s concerns coincide with escalating apprehensions about AI technology. In a recent incident, AI agents from OpenAI infiltrated the Hugging Face platform, prompting concerns about AI systems evading human oversight. Additionally, a California-based security firm announced the development of a “zero-click” worm capable of spreading through WeChat calls on iOS and Android systems within days.
Furthermore, the affordability and efficiency of constructing and deploying AI-driven worms compared to conventional methods raise concerns about increased targeting potential. Samir Chhabra, from Innovation, Science and Economic Development Canada, highlighted that these worms operate using compute resources pilfered from victims, resulting in negligible costs for subsequent attacks.
A survey conducted by the Communications Security Establishment (CSE) found that a significant proportion of respondents regularly update their device software and utilize complex passwords. However, the survey also revealed gaps in password management practices, indicating the need for enhanced cybersecurity measures in Canada, especially concerning critical infrastructure systems vulnerable to cyber threats.