Hackers linked to China allegedly used an autonomous artificial intelligence penetration testing tool named Cairn to compromise the servers of two Asian universities after discovering a zero-day vulnerability. Artificial intelligence is no longer limited to helping cybercriminals plan attacks, as it is now capable of independently identifying security flaws and infiltrating networks. During the campaign, AI agents found an unpatched weakness in an open-source medical imaging application that had no official advisory or security patch available at the time. Attackers exploited this flaw to gain access to one university system and subsequently used the same method to breach a second institution, underscoring the growing threat of automated AI-driven cyberattacks.
Cairn AI Tool Discovered on Public Server
Investigators uncovered the activity on September 10 when they found a publicly accessible server running an instance of Cairn, an open-source autonomous AI penetration testing framework. Inspection of the platform exposed its operational history, prior instructions, and related activities. The operator had entered prompts in Simplified Chinese, and the server timestamps were aligned with China Standard Time. Because the targets were universities located in East Asia, researchers assessed with confidence that the operation was linked to China, though the specific institutions and countries were not publicly named.
Cairn operates by deploying multiple AI agents simultaneously after receiving a target IP address and plain-language instructions. In this incident, the operator instructed the tool to obtain shell access to the targeted systems, which allows attackers to execute commands and carry out further unauthorized actions. Once activated, the AI agents independently mapped out potential attack paths, tested them, and adjusted their tactics based on the results without requiring step-by-step guidance from the human operator.
Persistent Automated Attempts and Zero-Day Exploitation
Operational logs revealed extensive automated probing by the AI tool. In one attempt, the agents executed 148 actions over approximately 16 hours without success, including a failed effort to exploit the known Log4Shell vulnerability. In another instance, the agents spent more than six hours attempting to breach a university server by testing over 30,000 password combinations associated with 80 valid usernames, which also failed. However, in a subsequent operation, the AI agents identified a path traversal flaw in an open-source medical imaging application and leveraged it to secure system access.
A path traversal vulnerability permits attackers to access files located outside an application’s intended directory structure. Because no security advisory, patch, or public record existed when the flaw was found, investigators classified it as a zero-day vulnerability. Attackers used this same weakness to compromise the network of a second university. Maintainers of the open-source medical imaging software were later notified and released a fix to resolve the security issue.
Security Agencies Warn Over Speed of AI Agents
Cybersecurity specialists expressed concern over the relentless pace and endurance of the AI agents, which operated continuously for 16 hours and adapted their methods after every failed attempt. Experts warned that if autonomous agents can discover and exploit zero-day vulnerabilities within minutes, traditional security operations centers may struggle to investigate alerts and respond in time. Amazon Web Services (AWS), working alongside a government partner, notified the affected organizations about the intrusion while keeping the identities of the universities confidential.
The incident adds to heightened scrutiny surrounding AI-powered offensive tools such as Artex and Cairn. Artex has previously been linked to cyberattacks targeting financial institutions in South Korea and Japan, prompting warnings from a United States investigative agency that such frameworks can automate operations from initial reconnaissance to data theft. Earlier reports regarding Cairn also indicated its use in campaigns targeting US aerospace firms and online retail businesses. Researchers noted that because Cairn is a general-purpose framework, its misuse extends beyond legitimate security testing and represents an escalating risk for organizations worldwide.
