AI Agents Powered by Cairn Were Used to Breach More Than 100 Websites

Threat actors are using an open-source AI orchestration platform to automate attacks against e-commerce and payment websites, including a campaign linked to compromises at more than 100 organizations.

NorForge discovered more than 20 publicly exposed Cairn control panels that appeared connected to malicious activity. The panels revealed targeting instructions, discovered vulnerabilities, stolen secrets and evidence of persistent access.

Cairn is a general-purpose problem-solving engine that coordinates AI agents toward a defined objective. In the observed campaigns, operators supplied targets and goals while agents investigated applications, selected attack paths and attempted exploitation.

One prolific cluster contained more than 1,100 projects. Independent reporting connected successful intrusions to more than 100 companies, including Deere, SharkNinja and Cummins. Deere said it found no indication that sensitive information was accessed, while SharkNinja said the affected site was isolated and the incident contained. Cummins said it was investigating a third-party website.

The research does not prove every project labeled successful resulted in a confirmed breach. It does, however, document how a single operator can run large numbers of long-duration intrusion attempts with limited direct interaction.

"A 10% success rate is a bad day for a human pen tester. Spread across thousands of automated attempts, it's a business model," said Justin Beals, CEO and founder of Strike Graph.

Defenders should reduce exposed administrative interfaces, rotate secrets found in web applications, monitor for unfamiliar files and web shells, and continuously test public-facing systems. AI agents are making persistent, large-scale probing less expensive for attackers.