Amazon’s threat intelligence division has identified a campaign in which Chinese-based actors utilized an autonomous AI penetration-testing framework known as Cairn to breach university networks in East Asia. The investigation, which commenced on September 10, began after security researchers discovered a publicly accessible server hosting an instance of the open-source tool. By reviewing the platform’s operational history, investigators observed prompts written in Simplified Chinese, configured to China Standard Time, which directed the AI to target specific IP addresses with the objective of obtaining shell access.
Cairn functions by coordinating multiple AI agents that autonomously plan, execute, and refine their attack strategies based on high-level, plain-language instructions. According to Amazon, this capability allows the software to operate continuously without human intervention. In one observed instance, the agents performed 148 distinct actions over a 16-hour period. While many of these attempts were unsuccessful—including a failed effort to exploit the well-known Log4Shell vulnerability and a six-hour campaign attempting to brute-force over 30,000 credential combinations—the agents eventually succeeded by identifying and exploiting a previously unknown vulnerability.
Amazon has confirmed that the successful breach involved a path traversal vulnerability found within an open-source medical imaging AI application. At the time of the incident, this flaw had no associated security advisory, patch, or mention in vulnerability databases. Based on this absence of documentation, Amazon’s threat intelligence team has assessed with high confidence that the discovery constituted a zero-day exploit. The attackers subsequently leveraged this same vulnerability to compromise a second university.
In response to the findings, Amazon coordinated with the maintainers of the medical imaging software to develop and release a fix for the vulnerability. Furthermore, the company collaborated with government partners to notify the affected academic institutions. While Amazon has confirmed the nature of the intrusion and the methodology used by the attackers, the company has declined to disclose the names of the universities or the specific countries in which they are located.
The use of Cairn highlights a growing trend in the cybersecurity landscape regarding the deployment of autonomous, multi-agent offensive frameworks. Cairn is categorized as a general-purpose tool, which some researchers argue makes it more unpredictable and versatile than other specialized hacking utilities. Because the software lacks internal structuring specifically designed for cybersecurity, it can adapt to various environments and tasks, making it a significant concern for traditional security operations centers that rely on manual triage.
This incident follows a series of warnings from various security entities regarding similar AI-driven threats. For instance, the FBI recently issued an advisory concerning a tool called Artex, which is capable of automating multiple phases of a cyber intrusion, from initial reconnaissance to data exfiltration. Artex has been linked to attacks on financial institutions in South Korea and Japan, as well as a supply chain attack involving the IT management software provider ConnectWise.
The broader ecosystem of offensive AI frameworks is expanding rapidly, with researchers noting the emergence of tools such as CyberStrikeAI and PentAGI. Reports indicate that Cairn has been observed in various sectors, including the U.S. aerospace industry and retail, where it has been employed by financially motivated hacking groups. The speed at which these AI agents operate poses a fundamental challenge to defenders, as the time required to discover and exploit a vulnerability can be significantly shorter than the time required for human analysts to identify and respond to an ongoing alert.






