Cybersecurity researchers have identified a large-scale campaign in which autonomous artificial intelligence agents were used to target more than 1,200 company websites, highlighting the growing risks associated with AI-assisted cyber activity. The campaign involved AI agents being directed toward websites with limited human involvement, allowing attackers to automate parts of the discovery and exploitation process. Security firm NorForge uncovered the activity after finding exposed control panels linked to Cairn, an open-source AI tool designed for authorised penetration testing but reportedly misused in this campaign. According to researchers, the activity involved attempts against websites belonging to more than 100 organisations, including companies such as Deere, SharkNinja and Cummins. The extent of any data access or theft remains unclear, while some organisations reportedly confirmed that they had taken steps to address the activity. Researchers found that attackers had left certain AI control panels accessible online, allowing investigators to observe their activities and analyse how autonomous agents were being used during the campaign.
NorForge researcher Snorre Fagerland discovered around two dozen exposed Cairn instances being used for suspicious operations. The tool allows AI agents to receive objectives and complete tasks step by step while sharing findings with human operators. Although Cairn was developed for legitimate security testing purposes, researchers said some operators used it to search for weaknesses in online platforms without authorisation. Cairn’s developer also stated that the tool was not created for harmful activities and that misuse was not intended. The investigation found that one major activity cluster, identified by NorForge as AIA0001, contained more than 1,100 attack projects connected to Chinese-language control panels. Researchers said the campaign appeared focused on maintaining access to websites and online systems for future use. In some cases, AI agents were instructed to identify ways to execute code on websites and deploy web shell uploaders, which can provide attackers with continued access if successfully installed. Researchers also identified cases involving weaknesses in outdated software components and unsafe configurations, although they noted that attack methods varied between victims.
The campaign demonstrates how AI tools can reduce the time and effort required for certain stages of cyber attacks. Researchers found that successful compromises took less than a day on average, while the estimated cost of individual attempts remained relatively low. NorForge estimated that the activity used thousands of dollars worth of AI model tokens, showing how affordable access to advanced AI capabilities could potentially enable more automated attack attempts. The investigation also highlighted risks created by exposed AI management panels. Researchers said these panels contained information about targeted websites, discovered weaknesses, credentials and other sensitive details. A security review using the Trufflehog secrets scanner reportedly found hundreds of exposed secrets, with some still appearing to be valid. Researchers warned that unsecured AI tools can create additional risks even when they are not directly involved in successful compromises.
Security teams are advised to review publicly accessible AI tool panels, strengthen authentication controls and monitor unusual upload functions or outdated website components. Researchers also recommended reviewing Magento environments and applying relevant security updates where required, while noting that no evidence confirmed the Magento vulnerability was used in the identified campaign. The findings add to growing concerns around the use of artificial intelligence in cybersecurity, as attackers and defenders continue to explore the capabilities of autonomous systems. While AI can support security testing and defence operations, researchers said organisations must ensure proper access controls, monitoring and secure deployment practices to prevent misuse of emerging technologies.
Follow the SPIN IDG WhatsApp Channel for updates across the Smart Pakistan Insights Network covering all of Pakistan’s technology ecosystem.
