Detect Deepfakesby Resemble AI
Agentic AI attack

University server agentic AI attack — Nov 2026

Suspected China-linked hackers used autonomous AI agents to exploit a critical path-traversal zero-day in MONAI Label software to gain root access to a.

Reported date
Nov 21, 2026
Target
university server
Agent type
Other AI agent
Agent role
Used by the attacker

The exact incident date was not established. This entry is dated by its source report.

Updated Nov 22, 2026 · 2 min read

What happened

Amazon Threat Intelligence reported that suspected China-linked actors utilized autonomous AI agents to exploit a previously unknown path-traversal vulnerability in MONAI Label, an open-source medical imaging server. The vulnerability, tracked as CVE-2026-133968, affects versions 0.8.5 and earlier. Because the software defaults to having authentication disabled, attackers were able to exploit the flaw without needing an account.

Investigators reconstructed the attack after discovering an exposed Cairn control panel, an open-source framework used to manage AI agents for automated penetration testing. The panel logs revealed that the agents were tasked with obtaining shell access. Within approximately 11 minutes, the agents mapped the application interface and began writing files to the server. While the initial file write occurred quickly, the agents later achieved root access in a separate session by planting a file that Python executes automatically at startup. By triggering a training feature, the agents forced the server to load the malicious file, establishing a root-level connection back to the attacker.

Amazon’s report suggests the agents operated with minimal step-by-step guidance, selecting their own methods after the operator defined the target and goal. The campaign was linked to a China-based actor with moderate confidence based on Simplified Chinese prompts, time zone settings, and the nature of the targets. While the agents were also provided access to a second university network, Amazon has not confirmed a successful compromise there.

As of the reporting date, the MONAI Label maintainers have merged a fix into the source code to prevent path traversal, but no official patched version has been released for installation. The vulnerability is rated as critical with a 9.8 CVSS score. Amazon noted that their findings are based on their own reconstruction and have not been independently verified. There is no evidence that patient records were accessed or that the attackers moved beyond the university targets. Security experts advise organizations to restrict network access to MONAI Label instances, enable authentication, and ensure the server runs under a limited-privilege account rather than root.

Evidence in the reporting

Incident evidence
gain root access to a university server
Agent involvement
the agents selected the methods themselves

Sources