A Security Researcher Finds Hackers Using AI Programs to Break Into Company Websites, at About $50 per Break-In
A security researcher reports watching a criminal group use AI programs to attack more than a thousand online shops and company websites, working on their own for hours at a time and costing only a few dollars per attempt. How much was stolen is unknown, and the companies named so far describe limited damage, but it shows break-ins that once needed skilled people being run cheaply by software.
A security research firm called NorForge published a report today describing criminals using AI programs to break into company websites, and watching them do it almost as it happened.
The tool they used is called Cairn. It is free and openly published, and it was built for legitimate security testing, where a company hires experts to try to break in so it can find and fix its weak spots. Cairn hands an AI program a target and a goal, then lets it work through the problem step by step on its own, keeping notes as it goes. Its author told reporters it "was not designed for hacking attacks."
NorForge's researcher, Snorre Fagerland, found the criminals because they had left Cairn's control screens open on the internet, where anyone could look. He found more than 20 that he judged were likely being used for crime and sorted them into three groups. The busiest, whose instructions were written in Chinese, ran more than 1,100 attacks, mostly on online shops and payment systems in the United States, Japan and Australia. The goal was usually to sneak in a hidden way back into the website for later use.
The numbers are rough, and NorForge says so: they "could be off by a clear margin." Its best guess is that about one attack in ten succeeded, that a successful one took a little under a day of the AI working on its own, and that each attempt cost about $5 in AI fees. Fagerland estimates the whole campaign cost around $5,000, or about $50 for each successful break-in. News reports say the attacks reached more than 100 companies, including the tractor maker Deere, which said it had no sign that sensitive information was accessed, and the appliance maker SharkNinja, which said only a separate product-support site was hit.
Much is still unknown. Nobody yet knows how much data, if any, was taken. The labels on the attackers' setup hint that they used a Chinese AI program driven by a coding tool from Anthropic, but neither NorForge nor reporters could confirm which AI programs were really involved. Security researchers have warned for months that AI could make hacking cheaper and faster. This report is one of the clearest looks yet at what that seems to mean in practice: one operator, a free tool, and a few thousand dollars.