Amazon says a Chinese-nexus actor's Cairn AI agents found a zero-day on their own; BleepingComputer names it MONAI Label

· CyberSecurity

TLDR

  • Amazon says a Chinese-nexus actor's Cairn AI agents found a zero-day in 11 minutes; BleepingComputer names it MONAI Label.13
  • Agents got root via a sitecustomize.py reverse shell; Amazon infers the actor then reused the flaw on a second university.1
  • MONAI Label's advisory lists no patched version for 0.8.5 and earlier, and authentication is off by default.6

Amazon Threat Intelligence says that on 10 September 2026 it identified a server with a publicly hosted instance of Cairn, an open-source autonomous AI penetration-testing framework, running without authentication.1 Because the operator left it open, the platform's operational history, including project configurations, actor prompts and AI output, was visible to anyone on the internet.1 The post, by CJ Moses, CISO of Amazon Integrated Security, says the prompts were in Simplified Chinese, script timestamps were set to China Standard Time, and the campaign targeted IP addresses associated with universities in East Asia outside mainland China for about 48 hours from 8 September.1 Amazon assesses with moderate confidence that the actor is Chinese-nexus, citing those prompts and timestamps and the targeting of universities likely of strategic interest to China.1 The actor gave the agents lists of IP addresses with variations of the goal "obtain shell access", and the agents planned, executed and adjusted without the operator choosing each next step.1

Most of the agents' attempts failed. Amazon describes roughly 16 hours and 148 sequential actions against a Redis instance configured to allow access without authentication, covering replication abuse, scripting sandbox escape, configuration file writes and memory corruption, plus a Log4Shell attempt against another service on the same system.1 After those failed, the agents mapped the internal Docker network through a server-side request forgery flaw in a co-located Qdrant instance and ran credential sprays against PostgreSQL, SSH and other services, with no success.1 A separate operation of more than 6 hours against a university GPU server built a list of about 80 real usernames from public sources and tested more than 30,000 credential combinations without finding a working login.1

Against a target running an open-source medical imaging AI application, the outcome was different. The agents identified the framework, mapped its API endpoints and found an upload endpoint with a filename parameter that was vulnerable to path traversal; in about 11 minutes they went from mapping the API to arbitrarily writing files on the server.1 In a later session they used the flaw to write a Python reverse shell payload to sitecustomize.py, which Python imports at interpreter startup, then triggered a training API endpoint to spawn a new Python process that established a reverse shell running as root.1 Amazon assesses with high confidence that it was a zero-day, based on the absence of any corresponding fix, issue, advisory or prior mention in vulnerability databases for the affected project.1 Amazon's post does not name the application; BleepingComputer identifies it as MONAI Label, which it describes as an open-source medical imaging application that uses AI to help researchers label medical images.3 In another session the actor gave the agents a proxy tunnelling into a second university network; the server behind it ran the same vulnerable software, and Amazon says it infers the actor had used the zero-day to compromise it.1 The Record reports that AWS worked with a government partner to notify the affected organisations and declined to name the universities or the country.2

The GitHub advisory for MONAI Label, GHSA-qc22-rmgc-vg94, was published on 18 November and rates the flaw critical.6 It says the PUT /datastore/image endpoint uses a caller-supplied image query parameter as the image ID without validation, so a value such as ../../../tmp/x lets an attacker write uploaded content anywhere the server process can write, which can lead to remote code execution.6 It lists versions up to 0.8.5 as affected and patched versions as "None", says the fix is merged in pull request #1915, and notes that authentication is off by default, so any client that can reach the server can exploit it; with authentication on, the ANNOTATOR role is enough.6 Its workarounds are to enable authentication and withhold the ANNOTATOR role from untrusted users, keep the server off untrusted networks, and run it as an unprivileged user with write access limited to the datastore directory.6 The advisory shows no CVE ID, while the GitHub Advisory Database entry published on 20 November lists CVE-2026-133968 with a CVSS 3.1 score of 9.8 and says the flaw was "exploited in the wild starting in September 2026".67 BleepingComputer says the maintainers released a security update on 18 November, which conflicts with the advisory's "None" and with VentureBeat; the date matches the advisory's publication, so it may refer to the advisory and the merged fix rather than a release.3 VentureBeat says the fix was merged on 19 October, that the newest release on PyPI is still 0.8.5, uploaded on 25 November 2024, and that the advisory says the fix will be included in the next release.4 AWS says the maintainers released a fix and a GitHub Security Advisory.1

Amazon lists two IP addresses in records dated 8 to 10 September: 142[.]4.98.145 and 142[.]4.107.34.1 It says Gambit Security documented a separate, financially motivated actor using Cairn against online retailers.1 The Register reports that a Gambit report documented a Chinese operator using Cairn in at least 105 attacks, compromising at least 27 companies "to varying degrees".5 Google Threat Intelligence Group's 16 November report lists CAIRN among open-source multi-agent frameworks it sees abused, and says it observed CAIRN used in intrusion activity against targets including the aerospace sector in the US.8 For defenders, Amazon says an agent that finds and exploits a zero-day in minutes finishes before many security operations centres can triage an alert, and advises prioritising high-confidence detections that trigger automated response.1

Why it matters: the agents went from mapping an API to arbitrary file writes in about 11 minutes, which Amazon says is faster than many security operations centres can triage an alert, and the advisory lists no patched MONAI Label version.16

Sources

  1. Chinese-nexus actor’s Cairn AI agents found a zero-day on their own (AWS Security Blog)
  2. Amazon says Chinese hackers used Cairn AI tool to discover apparent zero-day during attack on university (The Record)
  3. Hackers use autonomous AI tool Cairn to find zero-day, take over server (BleepingComputer)
  4. Amazon says attackers' AI agents found and exploited a zero-day in 11 minutes, but no patched release is available (VentureBeat)
  5. China attackers deployed AI agents that autonomously found and exploited a 0-day to take over a university server (The Register)
  6. Path traversal in datastore upload endpoints allows arbitrary file write (Project-MONAI (GitHub Security Advisory))
  7. monailabel (aka MONAI Label) through 0.8.5 allows... (CVE-2026-133968) (GitHub Advisory Database)
  8. GTIG AI Threat Tracker: Agentic Threat Activity Crosses a Significant Threshold (Google Cloud Blog)