Published: October 19, 2026 · Last updated: October 19, 2026
A cybersecurity researcher says a single hacker spent roughly $5,000 on AI agent tokens to break into more than 100 companies, including farm equipment maker Deere, home appliance brand SharkNinja and engine maker Cummins, paying as little as $50 per successful intrusion and letting the agents work largely unsupervised for up to 23 hours at a stretch.
Snorre Fagerland, a threat researcher at the cybersecurity firm NorForge who has worked in the field for nearly three decades, told Reuters he has never seen anything quite like it. Starting around September 25, a Chinese-speaking hacker set a small army of AI agents loose on e-commerce sites across the US, Canada, Japan and Australia. By last week, those agents had attempted to break into more than 1,200 sites and succeeded against over 100 of them.
The tool behind the campaign is called Cairn, an open-source AI platform that hands a target and a goal to a group of agents, which then work the problem step by step and post their progress to a shared board. Fagerland found roughly two dozen Cairn control panels left exposed on the open internet, more than 20 of which he judged malicious with high confidence. One panel tied to the Chinese-speaking cluster alone logged over 1,100 attack projects.
The instructions the hacker gave the agents were strikingly thin, sketched in fewer than 100 Chinese characters, according to Reuters. From there, the agents did the rest: probing public-facing applications, hunting through digital records for related infrastructure, finding a control panel, locating a subtle flaw in how it processed data and, after hours of trial and error, using that flaw to trick a server into running malicious commands. A successful break-in took about 23 hours on average.
Based on labels inside the Cairn data, the campaign appeared to run on a mix of Chinese AI lab DeepSeek’s models and Anthropic’s Claude Code, though Reuters could not confirm the exact combination. Fagerland’s rough estimate puts the hacker’s spending at about $5,000 in tokens total, or roughly $50 for every site actually breached. NorForge’s own report pegs the average closer to $5 per task once failed and abandoned attempts are counted in, underscoring just how cheap this kind of attack has become.
The goal, per NorForge, looks less like a smash-and-grab and more like planting a flag: agents that got in were tasked with installing a web shell, a foothold meant to give the hacker lasting access to shop and payment systems for later use. Deere told Reuters it had taken “necessary remediation actions” and had no indication sensitive information was accessed. SharkNinja said the breach hit a standalone support site with no sensitive data, and Cummins said it was investigating a potential attack on a site run by a third-party vendor. Two smaller businesses also confirmed they’d been compromised.
Fagerland’s dataset points to other operators running the same playbook. A separate English-language cluster told its agents to go after credit card numbers and cryptocurrency outright, with one hacker reportedly making off with a trove of card data and another trying to rob a crypto exchange directly. “It’s not script kiddie stuff,” Fagerland said. “The development in this field is astounding.”
None of this is happening in isolation. Reuters notes that a separate research firm, Gambit Security, found a Cairn-linked spree in September tied to the theft of at least 600,000 credit card records, and that cybersecurity firm Bitsight says chatter about autonomous hacking agents is surging across criminal forums. South Korea and Japan have already seen a string of breaches blamed on AI-wielding attackers this year.
What makes this case different is the price tag. Hiring a skilled human penetration tester for weeks of work costs real money and real time. Renting out a swarm of agents for $50 a hit does not. Fagerland said he had assumed, until recently, that agents like this would still need a human looking over their shoulder at each step. He said it’s now obvious that assumption no longer holds.
Related: DHS Orders DeepSeek Removed From US Federal Civilian IT Systems and Burger King Japan Data Breach Exposes 11 Million App Records.
Bottom Line: The part of this story that should worry security teams isn’t the $50 price tag, it’s the 23-hour autopilot. A hacker who used to need real skill to find a server misconfiguration can now describe the goal in a sentence and let a handful of cheap AI agents grind through the trial and error overnight. That math favors the attacker a lot more than it favors anyone defending a mid-size e-commerce site on a shoestring budget.
Follow Teck Hustlers: Facebook · X · Instagram · LinkedIn · TikTok · Pinterest
The Teck Hustlers Editorial Team covers technology, business, finance and gaming news for readers who want the story behind the headline. Every article is researched, written and edited by our in-house team, checked against company announcements, filings and reputable reporting, and updated when new facts emerge. We focus on what each development means for founders, professionals, investors and everyday users. Have a tip or a correction? Reach us through our contact page.
