Hacker Breaches 100 Organizations With AI Agents, Demonstrates a New Era of Black Hat Capability Has Arrived

New research from security firm NorForge indicates an era of cybersecurity that IT professionals have been dreading is now fully upon us: individual hackers are successfully deploying fully automated AI agents that handle every step of a breach from start to finish, and at relatively trivial cost.

The research tracks a number of clusters of threat actors that have left their activity in an open source penetration testing platform exposed, the largest being one that appears to work out of China and has attempted over 1,100 attacks thus far. About 100 of these have been successful, with the user giving the AI agents simple instructions at the outset and not intervening in any other way. They are also not spending a tremendous amount on these attacks, at an estimated cost of about $50 per successful breach.

AI agents able to uncover vulnerabilities, create complex attack chains without user monitoring

There has been great debate about if and when AI agents will be able to handle a breach from end to end on behalf of an attacker that does not necessarily know much (if anything) about hacking. Some believed this day was still years off, and others believed it might never come. This research indicates the capability is already within reach, though there are still some hurdles in terms of cost and skill.

The research is also clear that the AI agents are already far more than just a bunch of “script kiddies” shot out of a cannon; they probe each target relentlessly for novel vulnerabilities and come up with novel approaches based on what they find. All of this was done in this case without user input, and tended to take around 23 hours in the case of the successful breaches.

AI capability is not limited to nation-state actors

The key to these attacks is a set of open-source penetration testing platforms that are supposed to be used for legitimate purposes, but are clearly being abused by threat actors. The Cairn platform is not necessarily being abused more or less often than others, but has the unique quality of its users leaving their web control interfaces open to the internet fairly often (for unknown reasons).

The researchers simply trawled through these open panels to see what people were up to. It turned out at least 20 seemed to be devoted to acting as a harness to direct AI models in hacking campaigns. In the case of the successful Chinese group, DeepSeek 4.1 Flash was doing most of the work with Claude Code in a support role.

The 100 breaches the hackers racked up also were not limited to small and poorly defended targets. They managed to breach Fortune 500 businesses Cummins and John Deere and popular home appliance vendor SharkNinja among other confirmed incidents. It is still not clear exactly who all the victims were.

The one silver lining thus far is that none of the breached parties have reported serious damage as of yet (Cummins has told the media it is still investigating, however). However, the exposed Cairn panels also contained exposed credentials that were harvested by the hacking AI agents; the researchers found that about 11% were still functioning.

The news should put IT defense teams on immediate alert. The biggest takeaway is that hackers can try about 1,000 targets, land breaches at about 10% of them and pay only about $50 per successful breach on average, all using tools that are currently and freely available. While cost and setup will probably still screen out most amateurs, this is a powerful boost to cyber criminal gangs that already have infrastructure in place and resources on hand to invest. A vulnerability that likely would have taken a year or more for someone to trip over can now be expected to be sniffed out by probing AI agents in about a day.