Researchers at NorForge report that attackers compromised at least 17 internet-exposed panels for Cairn, an open-source system that lets artificial intelligence agents carry out penetration-testing tasks. Once inside an unprotected panel, an intruder could create projects and instruct the agents to run commands on the very infrastructure hosting them, turning a defensive tool into an attack platform.

How the intrusions worked

The reports stress that this was not a newly disclosed software vulnerability. The panels were reachable without proper authentication, so anyone who found them could view projects and start new tasks. Attackers then directed agents to gather environment details, listening ports and configuration files, and to hunt for stored secrets, including AI model API keys. Two activity clusters, tracked provisionally as AIA0009 and AIA0010, were identified. Researchers found similar wording in their instructions but no shared hard evidence that both clusters belong to the same operator. One cluster prioritised persistence and malware deployment; the other emphasised mapping internal services and scanning for further exposed panels.

The wider lesson

The campaigns intensified around 20 October, with earlier reconnaissance possibly starting around 10 October, and were reported on 27–28 October. The incident underlines a simple, often overlooked rule: any powerful automation panel must never be left open to the internet. Organisations using Cairn or similar agent systems should require authentication, restrict network access, rotate any credentials that may have been exposed, and audit task histories for unfamiliar projects. A tool designed to find weaknesses becomes a serious liability when its own front door is left unlocked. Sources: GBHackers and CybersecurityNews.