AI Hackers Left Their Own Dashboards Wide Open

A team of AI workers sharing one whiteboard

Security startup NorForge found more than 20 control panels for an AI hacking tool called Cairn sitting open on the public internet, with no password needed to read them. The Neuron reports that the panels appear to belong to criminals, and they offer a rare look inside how crooks are now using AI agents to break into real businesses.

Cairn works like a small crew sharing a whiteboard. A human gives the agents a target and a goal, such as breaking into one online store, and that job becomes a project. The agents post what they learn and what they plan to try next on the shared board, and because the board remembers everything, a single project can keep running for hours or even days.

The scale is what should worry site owners

One cluster of 14 panels, driven by instructions written in Chinese, held more than 1,100 attack projects aimed mostly at online stores in the United States, Japan, and Australia. The Rundown puts the same operation at more than 1,200 sites targeted, with one operator appearing to compromise more than 100 companies, including names reported by Reuters and The Verge such as Cummins, Deere, and SharkNinja.

NorForge estimates about 10 percent of projects hit their goal, and the successful ones in that cluster took just under 23 hours on average from start to finish. A typical attempt cost roughly 5 dollars in model fees once failures are counted, which works out to roughly 50 dollars in model fees per successful break in.

They verified at least one break in was real

Skepticism is fair here, because dashboards can exaggerate. NorForge checked one claimed break in at a large hardware parts vendor and found the back door, a hidden page that lets attackers upload files to a site server, still sitting live on the vendor site. That is not a simulation. That is a real hole left open in a real store.

The exposed panels also leaked the attackers own secrets. Scanning them turned up hundreds of exposed passwords and access keys, and 11 percent were still valid. One group even posted the login for its own staging server. The people threatening company security forgot to lock their own command center.

The holes were boring, and that is the point

The agents were not inventing science fiction exploits. Their work list was the boring stuff that site owners have been warned about for years:

• Login flaws and pages that skip the login check entirely
• File upload pages that let strangers place code on a server
• Old known holes that stay unpatched while owners stay busy
• Worker names in the panels suggest common coding agents powered by the cheap DeepSeek V4.1 Flash model, though NorForge calls that its reading and not a confirmed fact

What a Florida business should do this week

If you run a website, a store, or a client portal, treat this as a maintenance alarm. Update the site and its plugins, close upload and login holes, change any password that was ever shared loosely, and ask anyone who tests your security how they protect their own AI dashboards. NorForge says even apparently legitimate testing setups were left exposed, which can leak a client secrets.

Keep the numbers in proportion. NorForge sells tools for containing AI agents, its 10 percent figure is its own estimate, and success varies by group. But the direction is clear: old holes now get found faster, because software can work through more than a thousand projects without getting tired. Boring maintenance just became a real competitive edge.