Timeline

Amazon says China-linked attackers' Cairn agents found and used a zero-day

Agents told to 'obtain shell access' failed against Redis and GPU servers over nearly a day, then went from mapping a medical-imaging app to writing files in 11 minutes.

  • Security & misuse
  • Notable

Amazon Threat Intelligence published an investigation into a campaign in which AI agents running on Cairn, an open-source autonomous penetration-testing framework, found a previously unknown vulnerability in an open-source application and used it to take over a server. Amazon said it spotted a Cairn instance exposed without authentication on 10 September 2026, which let its analysts read the operator’s prompts, configuration and the agents’ output. The campaign ran for about 48 hours from 8 September against IP addresses linked to universities in East Asia outside mainland China.

The operator gave the agents lists of addresses and variations of one goal, “obtain shell access”, and the agents planned and adjusted their approach without the operator choosing each step. Amazon’s account includes failures. Over roughly 16 hours and 148 actions the agents tried a long list of known techniques against a Redis server with no authentication, and did not get in. Another operation ran for more than 6 hours against a university GPU server, built a list of about 80 usernames from public sources and tried more than 30,000 credential combinations, again without success. Amazon said the two together showed the agents “kept adapting for nearly a full day without human direction”.

Against a medical-imaging application, the outcome differed. The agents mapped its API, found an upload endpoint whose filename parameter allowed path traversal, and within about 11 minutes could write arbitrary files; in a later session they used it to obtain a root reverse shell. Amazon said it assessed “with high confidence” that this was a zero-day, because it found no matching fix, issue or advisory. After the agents found it, the operator supplied a proxy into a second university network that Amazon inferred had been compromised the same way, and directed the agents at further machines there. The maintainers of the project, whose security advisory names MONAI Label, released a fix.

Amazon assessed “with moderate confidence” that the actor is Chinese-nexus, citing Simplified Chinese prompts, China Standard Time timestamps and targets of likely strategic interest. The affected organisations were not AWS customers, so AWS worked with a government partner to notify them. The post adds to Unit 42’s report on a China-linked group using Cairn, and to Gambit Security’s separate report of a criminal actor using it against online retailers.