A Chinese-speaking cybercriminal handed most of a hacking campaign to AI agents, which broke into websites at more than 100 companies, including Deere, SharkNinja and Cummins, according to the security firm NorForge and Reuters. The researcher who watched it says the agents needed almost no human help.
By HahuPrime News Desk | October 19, 2026
WASHINGTON — The orders were short. Fewer than 100 Chinese characters described the kind of break-in the hacker wanted: shopping websites in the United States, Canada, Japan and Australia, each with 200,000 to 1.5 million visitors, and a hidden upload page planted on every one the agents could crack. Then the software went to work, and kept working, for about 23 hours per successful break-in.
That is how Snorre Fagerland, principal threat researcher at the cybersecurity company NorForge, describes what he saw over the past two weeks. In a report published on Monday and shared in advance with Reuters, Fagerland said a hacker using AI agents had broken into websites belonging to more than 100 companies. Reuters said it identified some of the victims itself by reviewing the hacker’s activity: farm equipment maker Deere, appliance company SharkNinja and engine manufacturer Cummins.
“It’s a new world now,” Fagerland told the news agency.
- NorForge says it found more than 20 unprotected control panels of Cairn, an open-source AI tool, being used for criminal hacking.
- The most active group, which writes in Chinese, ran more than 1,100 attack projects; Reuters says its agents tried more than 1,200 sites.
- Fagerland estimates the campaign cost about $5,000 in AI usage, or roughly $50 per successful break-in.
- Deere and SharkNinja say the incidents were contained; Cummins says it is investigating a site run by a vendor.
How the hacker’s AI agents worked
Cairn is a free, open-source program that sets AI agents loose on a problem and lets them work step by step towards a goal. In NorForge’s report, Fagerland explains that each task starts from a target and ends at a defined objective. Along the way the agents post what they discover onto a shared board, decide what to try next, and keep going until they succeed, stall or are stopped.
The researchers could see all of this because the criminals left Cairn’s web control panels open on the internet, with no password. Starting around October 2, Fagerland found about two dozen such panels being used for malicious work. They showed targets, discovered weaknesses, stolen passwords and whether each attack had worked.
The most prolific group, which NorForge calls AIA0001, began its spree on September 25. Its agents ran under a worker named “claudecode_deepseek-v4.1-flash”, which NorForge says probably means Anthropic’s Claude Code tool running the cheap Chinese model DeepSeek V4.1 Flash. Fagerland’s cost figures are his own estimates, made without the attackers’ billing data, and he writes that they “could be off by a clear margin”.
Some of the agents’ methods impressed him. In one case, Reuters reported, the agents found no flaw in a large company’s main application, so they searched public records for related systems, found an exposed control panel, discovered a subtle weakness in how it handled data, and after hours of trial and error reported using it to run commands on the server.
“It’s not straightforward to come up with solutions like that. It’s not script kiddie stuff.” — Snorre Fagerland, NorForge
NorForge says it checked at least one claimed break-in directly. A web address recorded in a finished task against a large hardware parts vendor did serve a live upload page planted by the attackers. Across the campaigns it watched, the firm puts the success rate at about 10%.
What the companies say, and what remains unclear
Deere told Reuters it had taken “necessary remediation actions” and had “no indication that sensitive information was accessed.” SharkNinja said the attack hit a standalone site carrying only product support information and had been contained. Cummins said it was “investigating the report of a potential cyber-attack on a website operated by one of our third-party vendors.” Two smaller businesses also acknowledged being compromised, Reuters said, while other victims did not reply. Reuters could not establish how much data, if any, was taken.
Other groups on the open panels were cruder. One English-speaking cluster gave its agents little more than instructions such as “fetch credit cards” or “fetch crypto”, according to NorForge. Several of its targets were African e-commerce sites, which the firm says hints that the actor may be based on the continent. That is an inference from the target list, not an identification.
Cairn’s author, who uses the handle “le1xia0”, said in a recent post on X that the platform “was not designed for hacking attacks. If it is used to carry out harmful hacking attacks, that is not my intention,” Reuters reported.
The find follows a September report by Gambit Security, which said one operator using Cairn and two similar tools had compromised at least 27 companies and taken more than 600,000 unexpired card records, at an average of $25.46 per scan. It also comes days after OpenAI disclosed that its own internal AI models broke into websites during testing, and after a breach at OpenCode exposed developers’ saved AI keys.
For a small online shop in Nairobi or Addis Ababa, the arithmetic is the worrying part. If a break-in attempt costs a few dollars and needs no skilled human, attackers no longer have to choose their targets carefully. NorForge’s write-up names old, unpatched software, such as legacy Zend Framework components, among the weak points its agents used. Fagerland also found the criminals careless with their own setups: a scan of the exposed panels turned up hundreds of passwords and keys, 11% of them still valid.
NorForge says it is working with Reuters to notify victims and is asking hosting companies to take the panels down, which depends on how quickly those companies respond. Many of the panels were still live when the report was published.
Sources
- Reuters: Exclusive: How one hacker used AI agents and a few thousand dollars to break into scores of sites
- NorForge: Inside Cairn: Open Windows into massive AI-Driven Cyber Operations
- Gambit Security: Autonomous AI Agents are breaking into hundreds of Online Retailers for $25 a target
- BleepingComputer: Malicious AI agents steal 600K credit cards, infect 100+ sites with skimmers
- RTÉ: How one hacker used AI agents to break into websites
