The hospital-software scare is smaller than the viral version, but the speed isn't: Amazon says AI agents went from probing a medical imaging tool to writing files on the server in about 11 minutes. If you ship open-source software with the login switched off by default, that's your timeline now.
Amazon Threat Intelligence says a China-linked operator used autonomous AI agents to find and exploit a previously unknown flaw in MONAI Label, an open-source server that research teams use to label medical scans and train imaging models, according to SecurityWeek and The Hacker News, which both reported on the findings this week. The campaign ran for about 48 hours in September, so what's new this week is the disclosure, not the attack.
It's worth being exact about what was hit, because the aggregator digests going around blurred it. The targets were IP addresses belonging to universities in East Asia outside mainland China, not hospitals, and there's no public report of patient records being touched. MONAI Label is research tooling that sits close to clinical work, which is why the story spread, but nothing published so far shows a hospital system was breached. The size of the compromise beyond the university servers hasn't been disclosed.
The flaw is a path traversal bug in the file upload function. According to The Hacker News, it lets someone with no account write files outside the server's storage folder, and it affects MONAI Label 0.8.5 and earlier in the default setup. Authentication is off by default, which SecurityWeek flagged as part of why the bug was so easy to reach.
Amazon's account is specific about the pace. In its words, quoted by SecurityWeek, "in about 11 minutes, the agents went from mapping the application's API to arbitrarily writing files." The operators then got root by planting Python startup files that ran when a training process started, a reverse shell included. The human set the goal, which was shell access, and picked the targets. The agents chose the method.
FBI and Google warn AI hacking tools like Artex now break into networks in hours Google's threat team named Artex, Cairn, HexStrike AI and PentAGI in a new report, while the FBI says Artex campaigns have hit at least ten sectors, including South Korean banks and ConnectWise customers. More than 1,600 control consoles sat exposed online in early November. - FBI warning about Artex AI hacking tools - AI agents breaking into networks within hours
Amazon says it has high confidence this was a true zero-day, because it found no matching fix, issue, advisory or earlier mention in vulnerability databases for the project. The Hacker News identified the agent framework as Cairn, an open-source attack tool. The tool isn't new. The operator and the target are.
On attribution, Amazon is cautious. It ties the campaign to a China-linked actor with moderate confidence, citing Simplified Chinese prompts, China Standard Time settings and targets likely to interest China. No named group has been attached to it.
Who owes whom a patch? #
This is where it gets uncomfortable for anyone who maintains software. According to The Hacker News, the MONAI Label maintainers merged a source fix on October 19, but no patched release exists yet, and the advisory lists no available update. SecurityWeek reports the developers published their advisory this week, rate the bug critical, and say the fix will ship in the next release. So a fix has been sitting in the code for more than a month while the vulnerable default version is the only one you can download.
Frankly, that gap is the story. MONAI Label is a volunteer-scale open-source project, not a vendor with a legal team and a patch cadence, and nobody can sensibly sue it. The institutions running it carry the risk, and most won't be watching a GitHub commit log. If an agent can map an API and land a file write in the time it takes to make a coffee, a month between fix and release is a long time to leave the front door open.
It also fits a pattern that healthcare security teams have been warning about. The FBI and Google warned earlier this month, as the American Hospital Association relayed on November 16, that attackers are using agentic AI platforms to move from initial targeting to privileged access in under three hours in some intrusions. Amazon's case stands out because the agents found the bug themselves rather than exploiting a known one.
For now, the practical advice is plain. If you run MONAI Label, turn authentication on, keep it off the public internet, and build from the October 19 source fix rather than waiting. If you sell software into hospitals or universities, assume your default settings will be probed by something that doesn't sleep. Neither the MONAI Label maintainers nor Amazon has said how many servers were compromised, and no patched release has been scheduled.
Two flaws in CreditNirvana's AI collections platform exposed 6.6 million accounts Researcher Eaton Zveare says CreditNirvana, owned by Perfios, left Aditya Birla Capital Finance call recordings and borrower files open through a hardcoded API key and a skipped token check. The vendor patched both bugs on June 2, but no one has said whether borrowers were told. - CreditNirvana vulnerabilities exposed 6.6 million accounts - CreditNirvana AI debt collection security flaws
Also read: South Korea's education minister quits days after AI cheating hit the national exam • Dell, HPE, Lenovo and Supermicro raise AI server prices up to 45% after signing • Inside the AI tools transforming US prison calls
This article is posted in Technology News, check it out for more related stories.
Join the discussion #
Open in the community → Almost there. Sign in and your reply posts straight away.