READ. SCROLL. LISTEN.

Unbiased headlines. Facts, not spin.

Unbiased Headlines is an independently owned news organization. We bring you the facts, not opinions, written fresh from 110+ named sources. Sources linked so you can verify it yourself.

← Back to headlines

Amazon says China-linked hackers used Cairn AI agents to find and exploit a zero-day in medical imaging software

Amazon says China-linked hackers used Cairn AI agents to find and exploit a zero-day in medical imaging software
Amazon Threat Intelligence says a suspected Chinese-nexus actor pointed the open-source Cairn framework at East Asian universities, and its AI agents found and exploited an unpatched flaw in MONAI Label to get root on a server. The agents failed in most of their attempts, but the one success took about 11 minutes. The maintainers shipped a security update Nov. 18, after the Sept. 8-10 intrusions.

Amazon says AI agents, not a human operator, found the hole.

According to a report released Friday, Nov. 20 by CJ Moses, CISO of Amazon Integrated Security, a suspected Chinese-nexus hacker gave the open-source Cairn framework a list of IP addresses and a plain-language goal: "obtain shell access." The agents did the rest. "The AI agents planned, executed, and adjusted their approach as they learned, without the operator choosing each next step," Moses said.

How the intrusion worked

Amazon's threat intelligence team found the campaign on Sept. 10, when it came across a publicly reachable server hosting a Cairn instance with no authentication. That exposed the operator's prompt history. The prompts were in Simplified Chinese, and script timestamps were set to China Standard Time. Targets were universities in East Asia, outside mainland China. The activity ran from Sept. 8 to 10.

The winning flaw was a path traversal bug in MONAI Label, an open-source, AI-assisted medical image labeling application. The agents identified the framework, mapped its API, and found an upload endpoint with a filename parameter. They tested it, and it was vulnerable.

It took about 11 minutes to go from mapping the API to writing arbitrary files on the server. In a later session, the agents wrote a Python reverse shell into sitecustomize.py, a file that loads automatically when Python starts. They then triggered an API endpoint and got root.

The operator next supplied a proxy into a second university network running the same software and directed the agents at other machines.

Amazon says it assesses "with high confidence" that this was a zero-day. At discovery there was no matching advisory, fix, issue or prior mention in vulnerability databases for the project. Amazon told the maintainers, who released a security update Nov. 18.

Most of it failed

The agents were not magic. Before the MONAI Label break-in, one effort ran 16 hours and 148 actions against a university's Redis server. It tried abusing Redis replication, escaping its scripting sandbox, writing to configuration files and corrupting memory. It also tried Log4Shell and a server-side request forgery attack against a Qdrant instance. None worked.

A separate operation spent more than six hours spraying over 30,000 credential combinations, built from 80 real usernames scraped from a university's public websites and papers, at a GPU server. That failed too.

"Neither operation succeeded, yet together they show that the agents kept adapting for nearly a full day without human direction," Moses wrote.

Amazon attributes the campaign to a Chinese-nexus actor with moderate confidence. It bases that on the prompt language, activity timestamps and target selection. That is an assessment, not a legal finding.

Speed is the problem

Moses argues that an agent which finds and exploits a zero-day in minutes "completes its activity faster than many security operations centers can manually triage an alert." The agents, he said, "ran continuously for 16 hours without breaks, sleep, or working-hour constraints." Amazon urges organizations to lean more on automated defenses than manual investigation.

The skeptical read comes from the facts above. Axios, in its own Nov. 18 reporting, described AI-enabled hacking as "still in its infancy" and noted that cyberattacks remain scary even without clear evidence of AI assistance. In this case, the agents failed in the Redis and GPU-server efforts and succeeded only once. The operator also left the control server open to the internet, which is how Amazon saw the whole thing.

Cairn is also a legitimate tool. It was built for penetration testing and defense research, not crime.

Not an isolated case

Gambit Security documented Cairn abuse in a September credit card theft operation. Google reported it was used in an intrusion at a U.S. aerospace organization.

Palo Alto Networks' Unit 42 separately tracks a China-nexus group it calls CL-STA-1238. It says the group paired Cairn with two Chinese open-source models, DeepSeek and Zhipu GLM, against at least 19 organizations. Targets included advanced semiconductor firms, the U.S. defense industry and civil-society groups.

Unit 42 says the attackers got in or exfiltrated data at 10 of them. They obtained dozens of valid credentials, created administrator accounts inside a production system, and exposed more than a million personal records at one organization.

"Cairn and other similar tooling enable automation, which means threat actors can hit more targets faster," said Andy Piazza, Unit 42's senior director of threat intelligence. "It also allows them to carry out technical tasks without having to learn the underlying technologies."

Unit 42 also said that neither Zhipu GLM nor DeepSeek refused when the attackers framed the work as a client-approved engagement.

The FBI's Nov. 13 FLASH alert covers a related framework, ARTEX. It says criminals have used ARTEX since at least August, with some intrusions going from initial targeting to privileged access in under three hours. Researchers counted over 1,600 publicly exposed ARTEX consoles in early November, even after the original developer pulled the code repository Oct. 8.

Google's Threat Intelligence Group says attackers are also using autonomous frameworks including ARTEX, CAIRN, CyberStrikeAI and PentAGI. It says they are shortening the gap between a patch's release and active exploitation of the flaw it fixes.

What is still unknown

AWS said it worked with a government partner to notify the affected organizations. It declined to name the universities or the country they are in.

The MONAI Label fix came Nov. 18, after the Sept. 8-10 attacks. Anyone running the application needs that update, and it is unclear how many installations remain exposed.

Sources used for this briefing

8 links from 8 independent newsrooms (syndicated copies of the same report count once)

These are the sources it draws on, linked so you can verify.

center-left
The RegisterChina attackers deployed AI agents that autonomously found and exploited a 0-day to take over a university server
unknown
The RecordAmazon says Chinese hackers used Cairn AI tool to discover apparent zero-day during attack on university
unknown
ChainCatcherAmazon claims that AI entities have discovered and exploited zero-day vulnerabilities to take over university servers
unknown
Four Week MBAFree AI Hacking Kits: Cairn Ran 25 of 25 Ops on Chinese AI
unknown
GabThe terrifying rise of AI hacking tools
unknown
Cybersecurity DiveChina-nexus actor uses AI tool in campaign targeting US defense, semiconductor industries
unknown
BleepingComputerHackers use autonomous AI tool Cairn to find zero-day, take over server
unknown
American Bazaar OnlineGoogle warns hackers are using AI agents to launch cyberattacks at machine speed

More coverage: Amazon and China