3 minute read
Burglary just went on sale
Remember when hacking a big company took a team of hoodie-wearing pros and months of planning? Yeah, that’s cute. According to Reuters, one Chinese-speaking hacker used AI agents to try breaking into more than 1,200 websites starting September 25. More than 100 of them fell. Reuters identified John Deere, SharkNinja, and Cummins among the victims. The estimated bill? About $5,000 in AI tokens. That’s roughly $50 per successful break-in.
The night shift never sleeps
The research comes from Snorre Fagerland at security firm NorForge, who published his report Monday. The hacker used Cairn, an open-source platform that coordinates a bunch of AI agents toward one goal. Give it a target and an objective, and it gets to work. The instructions were under 100 Chinese characters. That’s shorter than most of my grocery lists. Successful attacks took a little under 23 hours on average. No coffee breaks, no sick days, no complaining about the Wi-Fi.
Guess which AI was on the payroll
Every task ran on a worker labeled “claudecode_deepseek-v4.1-flash.” NorForge thinks that probably means Anthropic’s Claude Code running DeepSeek’s cheap V4.1 Flash model. So, a likely mashup of an American coding tool and a bargain Chinese brain. Nobody has confirmed it, and so far there’s no public comment from Anthropic or DeepSeek. Still, if you’ve been reading along here, Claude has had one heck of a month.
The victims say it’s fine. Mostly.
Deere says it took “necessary remediation actions” and has “no indication that sensitive information was accessed.” SharkNinja says only a standalone product support site got hit and the threat is contained. Cummins is investigating a site run by one of its vendors. Reuters couldn’t figure out how much data, if any, walked out the door. So, a lot of “we’re looking into it,” which is corporate for “please stop calling.”
The crooks forgot to lock their own door
Here’s my favorite part. How did anyone find out? The hackers left their own Cairn control panels sitting exposed on the open internet, and Fagerland found them by scanning. NorForge put it politely: “Securing their own infrastructure does not appear to be among their strengths.” The AI is smart. The humans running it, not so much. Sound familiar?
The tool maker shrugs
Cairn’s author, who posts as le1xia0, said on X that it “was not designed for hacking attacks,” adding, “If it is used to carry out harmful hacking attacks, that is not my intention.” Sure. Every crowbar company says the same thing.
Only 10%? Don’t relax yet
Before you breathe easy, NorForge says the average success rate was about 10%. Sounds low until you remember the agents don’t get tired, don’t get bored, and cost around five bucks a try. Ten percent of forever is a lot of break-ins.
So, what now?
Fagerland said he used to assume hacker agents would still need babysitting. Not anymore. “The development in this field is astounding,” he said. Security teams built their defenses around human attackers who sleep and make typos. That playbook just got old. Your move, IT department.
Sources: Reuters, NorForge, The Next Web, IBTimes UK, Gizmodo