A Chinese-speaking cybercriminal utilized autonomous AI agents to infiltrate more than 100 companies, including global brands like Deere and Cummins. the campaign targeted roughly 1,200 e-commerce sites across Japan, Australia, Canada, and the U.S., costing the attacker approximately $5,000 in AI tokens.
The $50 price tag for a corporate breach
The economic barrier to high-level cybercrime has collapsed. According to a report provided to Reuters by Snorre Fagerland, a principal threat researcher at NorForge, the hacker spent roughly $5,000 in AI tokens to execute this campaign. This breaks down to an estimated cost of only $50 per successful intrusion.
The efficiency of this approach is stark. The attacker used a mixture of Anthropic's Claude Code and China's DeepSeek to power the agents. These tools didn't just run scripts; they spent an average of 23 hours autonomously probing targets for vulnerabilities before successfully breaking in.
How Deere, SharkNinja, and Cummins became targets
The scale of the attack hit several industrial giants. Deere informed Reuters that it had taken remediation steps and found no evidence that sensitive data was accessed. SharkNinja reported that the breach was limited to a standalone site used for product support, while Cummins is currently investigating a potential attack on a website managed by a third-party vendor.
The sophistication of the AI agents allowed them to move beyond simple attacks. Snorre Fagerland of NorForge observed agents that, after failing to find a public weakness, independently searched digital records to find associated infrastructure and eventually tricked a server into exectuing malicious commands via a control panel.
Cairn, DeepSeek, and the automation of vulnerability hunting
The backbone of this operation was Cairn, an open-source AI problem-solving platform. While the creator of Cairn, known by the handle le1xia0, stated on X that the platform was not designed for hacking, the researcher found dozens of instances of Cairn being used for malicious activity.
This represents a shift from "script kiddie" attacks to autonomous exploitation. As reported by Reuters, the hacker provided terse instructions in fewer than 100 Chinese characters, leaving the AI agents to handle the complex trial-and-error process of discovering and leveraging subtle software weaknesses without human supervision.
From Cursor AI to the South Korean hacking wave
This incident is part of a broader global surge in agentic cybercrime. In August, the security firm Gambit Security discovered that Russian-speaking extortionists used SpaceX's Cursor AI tool to breach seven different organizations. Another AI-driven spree documented by Gambit in September resulted in 25 compromised companies and the theft of 600,000 credit card records.
Similar patterns have recently emerged in Japan and South Korea, where AI-using criminals have launched high-profile attacks. These events suggest that the cybercriminal underground is rapidly adopting autonomous agents to increase the speed and volume of their operations, as noted by the cyber intelligence company Bitsight.
The missing data on unauthorized extraction
Despite the confirmed breaches, a critical piece of the puzzle remains missing: the total volume of stolen data. While one observed hacker stole credit card information and another targeted a cryptocurrency exchange, it remains unclear exactly how much sensitive corporate intelligence was extracted from the 100+ breached companies.
Furthermore,the full extent of the "AI army's" reach is unknown, as several victims listed in the Cairn data have not responded to inquiries. this leaves a gap in understanding whether the $5,000 investment yielded a proportional financial return for the attacker through the sale of stolen data.
Comments 0