Suspected Chinese attackers used autonomous AI agents to find a zero-day in open source software and then use it to take full control of a university server, according to Amazon CISO CJ Moses. The AI agents zero-day discovery targeted MONAI Label, an open source medical imaging AI application, and no human directed the search.
Amazon told the project’s maintainers, who have released a fix and a security advisory.
How Amazon found the campaign
On September 10, Amazon’s Threat Intelligence Team came across a server with a publicly exposed instance of Cairn. That gave analysts a view into a campaign aimed at universities in East Asia, outside mainland China.
Cairn is an open source penetration-testing framework. It has grown popular with cybercriminals who use it to run multi-stage intrusions without a person at the keyboard.
The operation ran for 48 hours starting September 8. The prompts were written in Simplified Chinese, and timestamps in the executed scripts were set to China Standard Time. Those details point to a China-linked operator, though they are indicators rather than proof.
A day of failed attempts
The agents did not find the zero-day straight away. First they spent about 16 hours and 148 sequential actions on a university’s Redis server. They abused Redis’s replication feature to load unauthorized code, escaped its scripting sandbox, wrote to configuration files and corrupted memory. None of it worked.
They switched tactics. The agents tried a Log4Shell injection against another service, a server-side request forgery attack on a vulnerable Qdrant instance running alongside it, and credential sprays against PostgreSQL, SSH and other tools.
A separate operation ran for more than six hours against a university GPU server. It used another credential spray, with a list of 80 real user names collected from the school’s public websites, academic papers and other documents.
Both operations failed. Moses wrote that together they show the agents kept adapting for nearly a full day without human direction.
The MONAI Label break-in
Then the agents found a path traversal vulnerability in MONAI Label, and this time they got in. According to Moses, they identified the application’s framework, mapped its API endpoints and found an upload endpoint with a filename parameter. They tested that parameter for path traversal and found it vulnerable.
From mapping the API to writing arbitrary files on the server took about 11 minutes. The agents later abused the same flaw to run a Python reverse shell as root.
In a second session, a human operator stepped in after the discovery. The operator gave the agents a proxy that tunneled into another university’s network, where a server ran the same vulnerable code.
Not an isolated case
Amazon’s findings follow earlier work from Gambit Security’s threat intelligence team. That team documented a Chinese operator using Cairn in at least 105 attacks, compromising at least 27 companies to varying degrees. That campaign also took place in September.
In addition to Cairn, the attacker used two other open source AI toolkits. The stolen data included credit card records and other sensitive information, taken from a Fortune 500 hospitality company, a major US airline, a large private US industrial supplies distributor and a US online fashion retailer, among others.
Moses wrote that the cases together show the autonomous agent toolkit being adopted in both espionage and criminal operations.
What defenders should take from it
- Speed: the agents went from API mapping to arbitrary file writes in about 11 minutes.
- Persistence: failed attacks did not stop them. They moved to new targets and techniques.
- Exposure: the attacker left a Cairn instance publicly reachable, which is how defenders got a look inside.
Teams running MONAI Label should apply the maintainers’ fix and read the advisory. Anyone who exposes Redis, Qdrant, PostgreSQL or SSH to the internet should treat constant automated probing as the baseline, since an agent can keep trying for hours without tiring.
What happens next
Expect more of these cases to surface. Cairn is free and open source, and two separate September campaigns have already shown criminals and suspected state-linked operators using it. The AI agents zero-day finding in MONAI Label suggests the tooling can now handle vulnerability discovery as well as scripted exploitation, which shortens the time defenders have to patch.
Source: theregister