An alarming security breach involving autonomous AI agents has compromised company websites worldwide, targeting over 1,200 sites. These AI-driven attacks, reportedly orchestrated by a Chinese-speaking criminal, were closely monitored by a security researcher who discovered that the control panels were left exposed on the internet.
Scope of the Breaches
The AI agents attempted to infiltrate websites of more than 100 companies, including major firms like Deere, SharkNinja, and Cummins. According to data reviewed by Reuters, the extent of data theft remains unclear. The operation was detected by Snorre Fagerland, a researcher at security firm NorForge, who identified the attacks by scanning for open-source AI tool Cairn copies used maliciously online.
Throughout early October, about two dozen Cairn instances were identified, allowing the researcher to observe the attackers in real-time. Cairn, designed for legitimate purposes like penetration testing, was misused to target online platforms without authorization. The tool’s creator, known as ‘le1xia0,’ clarified that Cairn was not intended for such harmful activities.
Targets and Methods of Attack
NorForge categorized the exposed control panels into three clusters, with the largest, referred to as AIA0001, linked to over 1,100 attack projects. This cluster, primarily targeting the US, Japan, and Australia, focused on gaining persistent access to retail and payment systems.
In one detailed account, attackers leveraged vulnerabilities in a hardware vendor’s system, successfully installing a web shell for future exploitation. Despite the breaches, affected companies like Deere and SharkNinja reported limited damage and have taken measures to address vulnerabilities.
Operating Mechanisms and Risks
The operation was characterized by minimal supervision and low cost, with each successful breach costing approximately $50. The AI agents were directed by short instructions, written in Chinese, and achieved a success rate of about 10% in under 23 hours per task.
Exposed control panels present additional risks by revealing targets, vulnerabilities, and compromised credentials. The leaked data included hundreds of secrets, with a significant portion still valid. These panels not only pose a threat from criminal entities but also represent a risk for legitimate penetration-testing firms that inadvertently leave client data exposed.
Future Implications and Recommendations
Organizations are advised to secure exposed control panels, monitor for unauthorized upload forms, and ensure timely patching of software vulnerabilities such as those in Magento. NorForge, in collaboration with Reuters, is actively working to alert affected parties and encourage hosting providers to secure or remove exposed panels.
This incident underscores the growing threat of AI-driven cyber-attacks and highlights the need for enhanced security measures and vigilance in safeguarding digital infrastructure.
