A sophisticated cyberattack campaign utilizing an autonomous AI platform known as Cairn has successfully identified and exploited a zero-day vulnerability in a medical imaging application. The campaign, which occurred between September 8 and 10, 2026, primarily targeted educational institutions across East Asia. Researchers have attributed the activity to a Chinese-nexus threat actor, citing specific prompt language, operational timestamps, and the nature of the targeted organizations as the basis for this assessment.
Cairn is designed as a legitimate penetration testing framework that coordinates multiple AI agents to perform cyber-operations. While intended for research and defensive hardening, the platform has increasingly been observed in malicious contexts. Beyond the medical imaging breach, the tool has been linked to large-scale credit card theft operations and a separate intrusion attempt against a United States-based aerospace organization.
The recent campaign against medical research infrastructure involved the exploitation of a path traversal zero-day vulnerability within MONAI Label, an open-source application used for labeling medical images. According to security researchers who analyzed the incident, the AI agents autonomously mapped the application’s API and identified the flaw within 11 minutes. Once the vulnerability was confirmed, the agents successfully wrote a Python reverse shell to the sitecustomize.py file, which executes automatically upon Python startup. By triggering a specific API endpoint, the agents gained root-level access to the server.
The attackers’ methodology involved providing the AI agents with high-level objectives, such as obtaining shell access, rather than manual command-line intervention. In the case of the MONAI Label breach, once the zero-day was weaponized, the operators directed the agents to utilize a proxy to pivot into a secondary university network running the same vulnerable software. The vulnerability was subsequently reported to the maintainers of MONAI Label, who issued a security update to remediate the flaw on November 18, 2026.
Despite the success of the zero-day exploit, the AI agents were not universally effective. Security analysis revealed instances where the agents failed to compromise systems, including a 16-hour, 148-action attempt to seize control of a Redis server. In another instance, the agents spent six hours attempting to brute-force a university GPU server, testing over 30,000 credential combinations without success.
The investigation into these operations was made possible after the attackers left their Cairn instance publicly accessible without authentication, allowing researchers to observe the internal logic and command history of the agents. Two specific IP addresses, 142.4.98.145 and 142.4.107.34, were identified as being associated with the campaign during the September timeframe.
The emergence of autonomous tools like Cairn underscores a shift in the threat landscape, where AI agents can discover and weaponize vulnerabilities at a velocity that outpaces human security teams. Experts suggest that the speed of these automated attacks necessitates a move toward more robust, automated defensive systems rather than relying on traditional manual incident response. The incident serves as a clear indication that the barrier to entry for executing complex, multi-stage cyberattacks is being lowered by the integration of autonomous AI agents.






