Chinese-linked attackers used the autonomous AI agent platform “Cairn” to discover and exploit a zero-day vulnerability in MONAI Label, an open-source medical imaging application. According to Amazon Threat Intelligence, the attackers gave the agent a simple instruction to “obtain shell access,” after which it identified the flaw, gained the ability to write files to the server, and established a reverse shell with root access.
Amazon uncovered the campaign after finding Cairn instances exposed to the internet without authentication. Those exposed instances allowed researchers to observe the attackers’ activity and the prompts given to the AI agents.
Researchers said the campaign ran between September 8 and 10, 2026, and primarily targeted educational institutions in East Asia outside China. Based on the language used in captured prompts, activity timestamps, and the targets selected, Amazon assessed with moderate confidence that the operation was conducted by a Chinese-affiliated actor.
Cairn is a legitimate penetration-testing tool designed for research and defensive improvements. It can coordinate multiple AI agents during cyber operations.
Gambit Security documented an exploit involving Cairn in September as part of a large-scale credit card theft operation. Google was also recently reported as having been deployed in attacks against U.S. aerospace organizations.
Amazon’s investigation found that AI agents were not successful in every operation. In one instance, an agent spent 16 hours performing 148 actions while attempting to take control of a Redis server, but failed.
In another attack against a university GPU server, an agent tried more than 30,000 credential combinations over six hours but could not brute-force its way into the system.
AI agents discover and exploit a zero-day
According to Amazon’s investigation, Cairn’s AI agents found a zero-day path-traversal vulnerability in MONAI Label. The application uses artificial intelligence to help researchers label medical images.
The vulnerability allowed the agent to write files anywhere on the server. The agent mapped the application’s API and exploited the flaw in less than 11 minutes.
During later sessions, the agent used the vulnerability to create a Python reverse shell. It wrote a sitecustomize.py file, which Python automatically loads at startup, and used it to trigger an API endpoint that established root access to the server.
After finding the zero-day, the operator gave the agent access through a proxy to a second university network running the same vulnerable software. The agent was then instructed to target additional machines.
Amazon reported the vulnerability to the MONAI Label administrators, who released a security update scheduled to address the issue on November 18.
Indicators linked to the campaign
Amazon shared two IP addresses associated with the activity observed between September 8 and 10, 2026:
- 142[.]4.98.145
- 142[.]4.107.34
Amazon warned that AI agents may discover and exploit vulnerabilities faster than security teams can respond. The company urged organizations to place greater emphasis on automated defenses rather than relying solely on manual investigation.
Source: www.bleepingcomputer.com


