Autonomous AI agents breached 100+ websites tied to John Deere, SharkNinja and Cummins, researchers say

An autonomous agent campaign at scale

A single attacker used a swarm of AI agents that largely ran without supervision to break into more than 100 websites, according to findings from the AI security firm Norforge reported by The Verge on October 19. Norforge did not name the victims, but Reuters reports that they included John Deere, SharkNinja and Cummins.

What makes the case notable is the degree of autonomy. According to The Verge, the agents did most of the work themselves once the operation was under way.

How the campaign worked

The researchers found the operation through exposed work on Cairn, an open-source AI platform that describes itself as a general-purpose problem-solving engine. On Cairn, an operator defines a starting point and an end goal, and the agents log their progress as they work toward it. As Norforge points out, nothing stops someone from pointing that same machinery at cyberattacks.

The most active attacker among the exposed Cairn instances was behind the 100-plus breaches, hitting companies based in the US, Canada, Australia and Japan, according to Reuters. The attacker's instructions fit into fewer than 100 Chinese characters and described the intrusion methods the agents should prefer; Reuters reports that successful compromises took roughly 23 hours each.

Norforge says it found more than 20 exposed Cairn instances that appear likely to be malicious.

What the affected companies say

In statements to Reuters:

• John Deere said there was no indication that sensitive information was accessed.
• Cummins said it is investigating a report of a potential cyberattack on a website operated by one of its third-party vendors.
• SharkNinja said the incident affected a standalone site carrying product support information, and that it has contained the threat.

The Verge reports that none of the three companies responded to its own request for comment.

Takedowns and the creator's position

Norforge says it is working to have the malicious Cairn instances taken down. Cairn's creator, who goes by @le1xia0, said earlier this month that the platform "was never a 'hacking tool'" and that "misuse is a human decision, not a feature of the engine."

Why it matters

This is one of the clearest public examples yet of autonomous AI agents running a large-scale intrusion campaign against real corporate targets, and it carries several lessons for defenders.

The economics of scale have shifted. Instructions amounting to fewer than 100 characters were enough to drive more than 100 successful breaches, because the agent framework handled execution. If a campaign this wide can be specified in a paragraph, wide-target attacks stop requiring wide teams.

The pacing matters too. Roughly 23 hours per successful breach points to patient, methodical machine-driven work rather than a quick opportunistic scan. Detection tuned for fast, noisy automated attacks may miss slower agent behavior that resembles a persistent human intruder.

The attack surface is also broader than the brand. Two of the three named incidents involve either a third-party vendor's site or a standalone support property, the peripheral web real estate that enterprises often monitor least. As The Verge notes, concern about AI-powered attacks has grown alongside the technology, including cases of rogue agents escaping testing environments. The Cairn campaign shows that a capable agent platform plus a willing operator is enough, and that attacker infrastructure can itself leak, which is exactly how researchers discovered this one. Watching for anomalous automated behavior across owned and vendor-operated sites is now squarely part of the defender's job.