Skip to the content
AI and Cybersecurity Learning and Practice Center Test automation Open the Learning Center

IncidentAI-enabled attacks

Cairn panels left open show criminal crews running AI pentest agents against online shops and payment sites

HighNorForgeCairn

NorForge published on October 19, 2026 what it found in Cairn dashboards exposed on the internet, located with Shodan and Censys. Cairn runs each task as a project with a goal; agents post facts and intents to a shared board and operators add hints. NorForge assessed more than 20 panels as likely malicious and grouped them into three clusters. AIA0001, assessed as malicious with high confidence, ran more than 1,100 attack projects across 14 panels against e-commerce sites with 200,000 to 1.5 million visitors, mostly in the US, Japan and Australia, with Canada also listed; agents were told to look for login SQL injection, unauthenticated endpoints, file upload abuse, deserialization and authentication bypass, and to plant uploader web shells for later access. AIA0002, also high confidence, tasked agents to fetch credit cards and cryptocurrency from merchant payment, pharmaceutical and other commerce sites and to move funds between wallets. AIA0003, assessed with low confidence, shifted from "authorized assessment" wording to plain exploitation. One agent worker carried Playwright, PHPGGC, Nmap, ffuf, Amass and a proof of concept for Magento CVE-2026-75650, and its name suggests a Claude Code harness running DeepSeek v4.1 Flash. NorForge confirmed a live uploader shell on a hardware parts vendor's site and command execution on an Australian e-commerce server; it estimates about 5 dollars per task and a success rate around 10 percent, both uncertain. The panels also exposed hundreds of secrets and some legitimate red-team work. Dark Reading reported on October 20 that Reuters named John Deere, which took remedial action and saw no sign sensitive data was accessed, SharkNinja, which said a standalone product-support site was affected and contained, and Cummins, which is investigating a third-party vendor's website.

Replay the attack step by step in the Learning Center

How it happened

  1. Operators gave Cairn agents a target site and a goal; the agents searched for web flaws and reported findings to a shared board.
  2. Successful projects ended with an uploader web shell for later access, or with card and wallet data sent to separate hosts.
  3. Target lists were chosen by traffic, and browser automation checked whether shops could process orders.

Impact

A confirmed web shell on a commercial site, command execution on an Australian shop's server, and attempts on card data and cryptocurrency; the full damage is unknown, and Reuters named three affected companies.

The fix

Keep internet-facing shops patched and scanned continuously, look for upload forms and PHP files you did not deploy, rotate leaked keys first, and if you run an AI pentest tool for clients, never leave its panel open.

Lessons

  • Agents make ordinary web flaws on mid-sized sites worth attacking at scale, a few dollars per attempt.
  • An exposed control panel gives the whole operation away, the attackers' and a legitimate red team's alike.

Sources