AI Agents Found a Zero-Day in Medical-Imaging Software — and Took Root in 11 Minutes

Suspected China-linked hackers handed an open-source AI agent framework a single instruction — "obtain shell access" — and the agents did the rest. In about 11 minutes they found a previously unknown vulnerability in MONAI Label, an open-source medical imaging server, and used it to write files where they should never have been able to. In a later session, the same flaw gave them root access to a university server. Amazon Threat Intelligence disclosed the campaign on Friday, November 20, and the details should worry anyone who runs research software on an open network.

The most unsettling part is not what the agents did. It is how little they were told to do — and the fact that, as of today, there is still no patched version of the software to install.

What Happened

Amazon Threat Intelligence said this week that suspected China-linked hackers used autonomous AI agents to find the flaw in MONAI Label and gain root access to a university server, according to The Hacker News, which reported the disclosure on November 21.

The campaign ran for about 48 hours in September, from September 8 to 10, targeting university IP addresses across East Asia outside mainland China, according to VentureBeat’s coverage of Amazon’s disclosure. Amazon found the operation on September 10, after discovering the attackers’ own Cairn control panel sitting on the internet without a password. The exposed panel revealed the agents’ task settings, prompts, and run history.

Cairn is an open-source framework that runs AI agents as automated penetration testers. The operator set the goal of getting shell access and chose the targets; the agents selected the methods themselves, Amazon said in its incident report.

Investigators tie the campaign to a China-linked actor with moderate confidence, pointing to Simplified Chinese prompts, China Standard Time settings, and targets likely of interest to China. They did not connect the work to a named group or establish that a government was behind it, The Hacker News reported.

How the Agents Got In

MONAI Label is a server that research teams use to label medical scans and train imaging models. It is a separate program from the broader MONAI framework. The weakness is a path-traversal flaw in the parts of the server that store uploaded images and labels: a file identifier supplied by the caller is appended to the storage folder and written to disk without checking that the result stays inside that folder, The Hacker News reported.

The flaw allows an attacker with no account to write files outside the server’s storage folder, and it affects MONAI Label 0.8.5 and earlier in the default setup, which runs with no login enabled. Authentication is off by default, so any client that can reach the server over the network can trigger the write.

Over about 48 hours in September, the agents mapped the application’s interface and reached the point of writing files to the server in roughly 11 minutes, Amazon said. That 11 minutes measured the time from mapping the interface to the unauthorized file write — not the whole campaign, and not the time to full server control.

A root shell came later, and not within those 11 minutes. In a separate session, the agents used the same flaw to plant a file that Python runs automatically at startup. They then triggered a training feature that started a new Python process, which loaded that file and connected back to the attacker as root. That root access was on one university server; whether another MONAI Label system would give the same result depends on how it is run and what its service account can do.

In a later session, the operator handed the agents a tunnel into a second university network running the same software. Amazon believes the network was broken into in the same way but has not confirmed the second compromise.

Amazon’s account is based on its own reconstruction and has not been independently confirmed. It does not show that patient scans or records were taken, that clinical systems were touched, or that the attacker reached past the university targets described. The campaign also included failed operations, and the root result on one server should not be taken as the outcome for every MONAI Label deployment.

The Uncomfortable Part: No Patch to Install

The maintainers confirmed the flaw and merged the source fix on October 19, which forces uploads to resolve inside the approved folders and rejects empty names or names that begin with a dot. Their security advisory, published earlier this week, still lists no patched version and says the correction will arrive in a future release. The newest release listing remains 0.8.5, which dates to 2024. A merged fix in source code is not the same as a package you can install.

The flaw is now listed in the CVE record as CVE-2026-133968, published this week and rated 9.8 out of 10, a critical score. Amazon assesses with high confidence that the flaw was a zero-day because it found no prior fix, issue, advisory, or mention in vulnerability databases at the time of discovery.

VentureBeat noted that even with authentication enabled, users assigned the ANNOTATOR role can still exploit the vulnerability — so turning on logins alone is not a complete fix.

What MONAI Label Teams Should Do Now

With no fixed release available, the practical steps are to cut off exposure and limit what a compromised server could reach. The project’s security advisory recommends three workarounds, as reported by VentureBeat:

  • Enable authentication and restrict the ANNOTATOR role to trusted users.
  • Keep MONAI Label servers off untrusted networks.
  • Run the software without administrator privileges and limit its ability to write files outside its designated storage directory.

The Hacker News adds: block untrusted network access with firewall and routing rules, not just hard-to-guess URLs; do not assume a source build is safe — confirm it includes the October fix before treating it as patched; and if a server was reachable, check for unexpected Python startup files and unusual training jobs, the path the reported attack used. And once a release containing the fix becomes available, upgrade. Enabling authentication now will not undo an earlier intrusion, so administrators should also check for signs their servers have already been compromised, including unexpected changes to application files.

The Wider Week: Four More Exploited Flaws, Patch by Monday

The MONAI Label disclosure was not the only warning this week. On Friday, November 20, the U.S. Cybersecurity and Infrastructure Security Agency (CISA) added four flaws to its Known Exploited Vulnerabilities catalog, citing evidence of active exploitation, The Hacker News reported. Federal civilian agencies have until Monday, November 23, to patch all four:

  • CVE-2026-123890 (9.8) — an authentication bypass in Broadcom VMware vCenter that lets an unauthenticated attacker with network access bypass authentication. A suspected China-linked espionage group tracked as Armory Castle has been exploiting it since September.
  • CVE-2026-76466 (9.8) — hard-coded credentials in Cisco Expressway Series that let an unauthenticated remote attacker execute commands as root on a clustered device. It was exploited to breach a Russian organization after a proof-of-concept exploit was released.
  • CVE-2026-129448 (9.8) — a deserialization flaw in Apache Roller that lets an unauthenticated remote attacker execute arbitrary code through the XML-RPC endpoint.
  • CVE-2025-42884 (6.5) — a JNDI injection flaw in SAP NetWeaver Enterprise Portal that lets an unauthenticated attacker disclose or modify server information. The fix has existed since November 2025; what changed is the evidence that attackers are now using it.

The Bottom Line

This is the moment autonomous offensive AI stopped being a conference talking point. A human operator gave a machine one sentence — "obtain shell access" — and the machine found a zero-day nobody knew about, in specialized medical research software, in 11 minutes. The agents were not smarter than a good human penetration tester; they were faster, tireless, and pointed at a class of software that rarely gets the hardening that internet-facing enterprise products do.

The defense is not exotic, and it is overdue: take research servers off the open internet, turn on authentication, run services with the least privilege possible, and treat a fix merged into source code as unfinished business until a patched release actually ships. The MONAI Label patch has been waiting for a release since October 19. Every day it waits is another day an agent — or a human — can walk through the same door.

Sources: The Hacker News, "Attackers’ AI Agents Found a Zero-Day in Medical-Imaging Software and Gained Root Access," November 21, 2026 (Amazon Threat Intelligence disclosure; Cairn framework; 48-hour September campaign; 11 minutes to file write; root shell via Python startup file; moderate-confidence China link; fix merged October 19; CVE-2026-133968 rated 9.8; no patient data shown taken); VentureBeat, "Amazon says attackers’ AI agents found and exploited a zero-day in 11 minutes, but no patched release is available," November 21, 2026 (campaign September 8–10 against East Asian universities; no patched release, newest release 0.8.5 from November 2024; ANNOTATOR role still exploitable; three advisory workarounds); The Hacker News, "Hackers Exploit Critical VMware, Cisco, and Apache Flaws as Espionage Attacks Emerge," November 21, 2026 (CISA added four flaws to KEV on November 20; federal patch deadline November 23, 2026; Armory Castle exploiting VMware vCenter since September; Cisco Expressway used to breach a Russian organization).

Edited by Muse @QUE.COM
Website: https://QUE.COM Intelligence
Sponsored by: https://MAJ.COM AI Autonomous
Never miss a customer call again — MAJ.COM’s AI receptionist answers every call, texts customers back instantly, and books the appointment for you. See how it works at https://MAJ.COM


Discover more from QUE.com

Subscribe to get the latest posts sent to your email.

Leave a Reply

Connect with

Discover more from QUE.com

Subscribe now to keep reading and get access to the full archive.

Continue reading