Skip to content
feed: live
0dayNews
ai tools

Cairn AI agents found a MONAI Label zero-day on their own

Amazon says a China-nexus actor told Cairn agents to get shell access, and they found and used a MONAI Label path traversal flaw, patched November 18.

Cairn AI agents found a MONAI Label zero-day on their own
Image: AI-generated — conceptual illustration, not incident evidence / 0dayNews AI Cover · Generated on-site infrastructure — no external license
kilobaudDave "Kilobaud" Ferris·Published ·2 min read

Amazon Threat Intelligence has described a campaign in which suspected Chinese-nexus attackers pointed a multi-agent platform called Cairn at a target with the instruction to “obtain shell access,” and the agents found and exploited a previously unknown flaw without further direction, according to BleepingComputer’s report.

What to do: if you run MONAI Label, update to the release the maintainers shipped on November 18. No CVE identifier appears in the reporting we reviewed, so we are not assigning one here.

What happened

The activity ran between September 8 and 10, 2026, and mostly targeted educational institutions in East Asia outside China. Amazon found it because the operators left their Cairn instance reachable without authentication, which exposed the operation itself.

The zero-day was a path traversal bug in MONAI Label, an open-source application that uses AI to help researchers label medical images. The flaw let the agents write files to arbitrary locations on the server. Amazon reports that about 11 minutes passed between the agents mapping the application’s API and exploiting the flaw. In a later session they used the write primitive to gain root on the server. We are leaving out the mechanics beyond that.

The operator then gave the agents a proxy into a second university network running the same software and told them to go after other machines. Amazon reported the bug to the maintainers, who released a fix on November 18.

The failures matter too

The agents were not uniformly effective. A 16-hour, 148-action attempt against a Redis server failed. Against a university GPU server, they tried more than 30,000 credential combinations over six hours and still did not get in. The successful case was the one where an application had a real, findable bug.

Attribution is to a Chinese-nexus actor with moderate confidence, based on the language of the recovered prompts, activity timestamps, and target choice.

Cairn is a legitimate tool

Cairn is a pentesting platform built for research and defense, and the same reporting notes earlier abuse. Gambit Security documented it in a credit card theft operation in September, and Google reported it used against a U.S. aerospace organization.

Analysis: the notable part is not any single technique. It is that the discovery step, which used to need a researcher with time and skill, was delegated by a one-line instruction. Amazon’s own advice is to lean more on automated defenses than manual investigation, since agents can find and use bugs faster than teams can triage. Research-grade software like MONAI Label, often deployed on university servers with little hardening, is the likeliest place to feel that shift first.

Amazon shared two associated IP addresses in its report, 142[.]4.98.145 and 142[.]4.107.34, for defenders who want to check logs.

Found this useful? Share it.