Exposed control panels show criminals using AI agents to break into online shops

Security firm NorForge published research on October 19, 2026 showing how criminals run AI agents to hack websites. It found more than 20 publicly reachable control panels for Cairn, an open-source agent tool that works step by step from a starting point toward a goal, that were likely being used for attacks. Many had no login protection at all.

The busiest group, which NorForge calls AIA0001, writes in Chinese and ran over 1,100 attack projects across 14 panels. It targeted e-commerce and payment sites, mostly in the US, Japan and Australia, with the goal of planting a hidden web shell for lasting access. The agents ran largely on their own, used a worker built around the cheap DeepSeek V4.1 Flash model, and took a little under 23 hours on average to reach a goal. NorForge estimates a typical task cost around five dollars. It puts the overall success rate near 10%. A second, English-speaking cluster focused on stealing card and crypto data.

The open panels also leaked the attackers' own secrets: NorForge ran a scanner over the data and found hundreds of credentials, 11% of them still valid. Separately, Reuters reported that victims of one operator included John Deere, SharkNinja and Cummins. John Deere and SharkNinja said no sensitive data was reached, while Cummins said it was investigating. NorForge is working with Reuters to notify victims and get the panels taken down.

Why it matters

This is a concrete look at autonomous agents carrying out real break-ins cheaply and at scale, and a warning that defenders now face attackers who can run many AI-driven intrusions at once.

Source: NorForge