
Amazon Threat Intelligence says attackers used an open-source autonomous AI pentesting framework called Cairn to find and exploit a previously unknown flaw in a medical imaging application, with no human choosing each step. Amazon found the campaign on September 10 when the attackers left their Cairn server reachable from the internet, exposing its prompts and history. The operator wrote prompts in Simplified Chinese, and the targets were universities in East Asia. Amazon assesses with moderate confidence that a Chinese-nexus actor was behind it, and with high confidence that the flaw was a zero-day because it found no matching fix, issue or advisory at the time.
Most of what the agents tried failed. One effort against an open Redis server ran about 16 hours and 148 actions without success, including a Log4Shell attempt, and another tried more than 30,000 credential combinations against about 80 usernames on a university GPU server without finding a working login. The exception was MONAI Label, an open-source medical image annotation tool. Amazon says the agents mapped its API, found an upload endpoint with a vulnerable filename parameter and, in about 11 minutes, could write files anywhere on the server. In a later session they wrote a Python reverse shell to sitecustomize.py, triggered a training endpoint to start a new Python process, and got a root shell. In another session the operator gave the agents a proxy into a second university network running the same software, which Amazon infers had been compromised with the same flaw. BleepingComputer, citing the report, says Gambit Security has documented Cairn in a credit-card theft operation and Google has reported it in an attack on an aerospace organization; those two reports were not read for this post.
How to check if you’re affected
Affected versions are MONAI Label releases up to and including 0.8.5 (the monailabel pip package), per the GitHub advisory GHSA-qc22-rmgc-vg94, which rates the flaw Critical and says there is no CVE. The advisory names PUT /datastore/image and PUT /datastore/label as the vulnerable endpoints. Authentication is off by default (MONAI_LABEL_AUTH_ENABLE=False), so any client that can reach the server can write files; with authentication on, the ANNOTATOR role is enough.
The advisory was published November 18, lists “Patched versions: None” and says the fix (pull request #1915) will be in the next MONAI Label release. BleepingComputer wrote that the maintainers “released a security update” on November 18; the advisory itself shows the fix in the source repository but no released version, so check the project’s releases before assuming a patched package exists. Until then, the advisory’s workarounds are:
- Turn on authentication (
MONAI_LABEL_AUTH_ENABLE=True) and do not give the ANNOTATOR role to untrusted users. - Do not expose the MONAI Label server to untrusted networks.
- Run the server as an unprivileged user whose write access is limited to the datastore directory.
Amazon published two IP addresses seen in records dated September 8 to 10, 2026: 142[.]4.98.145 and 142[.]4.107.34. Searching your logs for them is a reasonable check. Looking for an unexpected sitecustomize.py in the Python environment that runs the server follows from the technique Amazon described, but neither Amazon nor the advisory lists it as a check.
Amazon’s own conclusion is that AI agents can find and exploit flaws faster than many security teams can triage an alert, and it urges organizations to rely on high-confidence detections that trigger automated response, with people overseeing outcomes. Related earlier coverage of AI-driven intrusion tooling: FBI Flash Alert Warns of Artex AI Tool Behind Fast Intrusions.
Sources
- Amazon Security Blog: Chinese-nexus actor’s Cairn AI agents found a zero-day on their own
- GitHub Security Advisory GHSA-qc22-rmgc-vg94: Path traversal in datastore upload endpoints (MONAI Label)
- BleepingComputer: Hackers use autonomous AI tool Cairn to find zero-day, take over server
- The Record: Amazon says Chinese hackers used Cairn AI tool to discover apparent zero-day during attack on university
