AI Agents Hack More Than 100 Websites, Including Australian Targets, for Just US$5,000

Concerns around AI agents and our data security don’t seem to be going away. After OpenAI revealed several instances where AI agents either accessed crypto wallets without permission or entered government websites without authorisation, a major hacking campaign is now drawing attention.

A Reuters investigation revealed that a hacker had compromised websites belonging to more than 100 companies after attempting over 1,200 intrusions. The original findings, published by cybersecurity firm NorForge, didn’t name the victims, but further reporting by Reuters identified high-profile firms such as tractor-maker Deere, engine manufacturer Cummins and SharkNinja, the well-known home appliance producer.

According to NorForge, the Chinese-speaking hacker relied heavily on AI agents and did little of the work himself. The attacks, which began in late September, targeted companies in the US, Canada, Japan and Australia, with the agents independently searching for weaknesses in websites and attempting to gain access.

The hacker used an AI-powered platform called Cairn, which allows users to assign tasks to autonomous agents. Once given a target and instructions, the system can explore possible vulnerabilities and develop ways to exploit them with little human intervention.

What makes the campaign particularly concerning is its relatively low cost. According to NorForge researcher Snorre Fagerland, the operation may have cost as little as US$5,000 (AU$7,160) in AI processing expenses, although the estimate comes with considerable uncertainty.

The researcher estimated that around 10% of the observed attacks achieved their objectives, with successful attempts taking approximately 23 hours on average.

Related: Cysic Suspends Bridge After Attacker Mints 10 Million Unauthorised Tokens

Fagerland, who has almost three decades of experience in cybersecurity, told Reuters that some of the methods employed by the agents were surprisingly sophisticated.

Australian Websites Among Targets

Australian businesses were among those targeted, with NorForge documenting one case in which an AI agent successfully gained the ability to execute commands on an Australian e-commerce website’s server.

However, the full extent of the damage remains unclear. Reuters said it couldn’t establish whether the intrusions resulted in stolen data. Deere and SharkNinja confirmed that they had addressed security incidents, while Cummins said it was investigating a possible attack involving a third-party website.

Crypto Also in the Crosshairs

The investigation also uncovered another group using AI agents specifically for financial theft, including attempts to steal cryptocurrencies and credit card information.

Unlike the Chinese-speaking hacker, whose main objective appeared to be gaining access to websites for possible later exploitation, this separate group was more directly focused on obtaining financial assets.

NorForge found that the agents were instructed to retrieve crypto and payment information, sometimes with little more than a basic objective to guide them. In some cases, the technology was also used to transfer funds between crypto wallets.

The findings raise further concerns about the use of autonomous AI systems in cybercrime, particularly as increasingly sophisticated attacks can apparently be carried out at relatively little cost and with limited human involvement.

Related: Crypto Security Firm Says US$1.18 Billion Was Stolen in Third Quarter