Amazon says hackers used AI tool Cairn to find and exploit a zero-day
Amazon Threat Intelligence says AI agents found an unknown flaw in a medical imaging app and used it to take over a university server in East Asia, then breach a second university.
Key takeaways
- Amazon Threat Intelligence says attackers used the open-source Cairn framework to find and exploit an unknown flaw in MONAI Label.
- The AI agents went from mapping the app's API to exploiting the path traversal bug in 11 minutes, BleepingComputer reports.
- Amazon ties the September 8 to 10 attacks on East Asian universities to a Chinese-nexus actor with moderate confidence.
- Other attempts failed, including a 148-action effort over 16 hours and more than 30,000 credential combinations.
- MONAI Label's maintainers released a fix on November 18 after Amazon reported the flaw.
Attackers used Cairn, an open-source framework that coordinates several AI agents, to discover a previously unknown flaw in a medical imaging application and use it to take control of a university server, according to a report from Amazon Threat Intelligence covered by The Record and BleepingComputer. Amazon links the activity to a Chinese-nexus threat actor with moderate confidence.
Amazon found the campaign on September 10, after spotting a Cairn instance the operator had left reachable on the internet, said CJ Moses, CISO of Amazon Integrated Security, according to The Record. BleepingComputer reports the panel needed no authentication, which let researchers see the operation's history, including earlier prompts.
How did the AI agents get in?
The operator gave the agents IP addresses and the goal to "obtain shell access," The Record and BleepingComputer report. "Based on those inputs, the AI agents planned, executed, and adjusted their approach as they learned, without the operator choosing each next step," Moses said, according to The Record.
According to BleepingComputer, the agents found a path traversal bug in MONAI Label, an open-source tool that helps researchers label medical images with AI. The flaw let them write files anywhere on the server, and they moved from mapping the application's API to exploiting it in 11 minutes. In a later session they placed a Python reverse shell in the sitecustomize.py file and gained root access, BleepingComputer reports.
Amazon says it has high confidence the flaw was a zero-day, because it found no matching fix, issue, advisory or earlier mention in vulnerability databases, Moses said according to The Record. The attacker then used the same flaw to breach a second university, The Record reports. BleepingComputer says the operator then routed the agents through a proxy into another university's network, which ran the same vulnerable software, and pointed them at other machines there.
What did not work?
Not every attempt succeeded. One effort of 148 actions over 16 hours failed, including a try at the Log4Shell vulnerability, The Record reports; BleepingComputer says that effort targeted a Redis server. In another, the agents spent more than six hours trying more than 30,000 username and password pairs for 80 real accounts without getting in, according to The Record.
Who was targeted and what has been fixed?
The targets were mainly universities in East Asia outside China, and the attacks ran from September 8 to 10, according to BleepingComputer. Amazon based its attribution on the Simplified Chinese prompts, timestamps set to China Standard Time and the choice of targets, the outlets report. AWS declined to name the universities or their country, The Record reports, and said it worked with a government partner to notify them.
Amazon told the MONAI Label maintainers about the flaw, and a security update was released on November 18, according to BleepingComputer.
Moses said the speed of the agents stood out, noting that they "ran continuously for 16 hours without breaks, sleep, or working-hour constraints," according to The Record. Amazon urged organizations to lean more on automated defenses, BleepingComputer reports.
Is Cairn a hacking tool?
Cairn was built as a legitimate penetration-testing tool for research and defense, BleepingComputer notes. Gambit Security documented its use by a financially motivated group in September, and Google said this week it had seen Cairn in intrusions in the U.S. aerospace sector, The Record reports; BleepingComputer describes an attack on one U.S. aerospace organization. Gambit Security's threat intelligence director, Eyal Sela, told The Record that Cairn matters more than Artex, another AI hacking tool the FBI warned about last week, because Cairn is general-purpose.
Questions readers are asking
What is Cairn?
An open-source framework that coordinates multiple AI agents for penetration testing. It was built for research and defense, but attackers have used it, according to BleepingComputer and The Record.
Which software had the flaw?
MONAI Label, an open-source medical imaging application. The path traversal bug let the agents write files anywhere on the server, BleepingComputer reports.
Who was behind the attacks?
Amazon attributes them, with moderate confidence, to a Chinese-nexus threat actor. It pointed to the prompt language, the timestamps and the targets chosen.
Has the flaw been fixed?
Yes. The maintainers released a security update on November 18 after Amazon reported it, according to BleepingComputer.
Sources
- #amazon
- #cairn
- #ai-agents
- #zero-day
- #monai-label
- #threat-intelligence
Get the day's top AI & tech stories on WhatsApp: Follow the NextFish channel
Get stories like this every morning.
Related
Cybersecurity2 min
Supplier hack exposes payee data at more than 40 Irish credit unions
Belfast-based Wellington IT's mobile banking platform was breached in early October; credit unions were told in mid-November. Officials see no evidence funds were accessed.
Cybersecurity2 min
Chase, Visa and Mastercard confirm incidents at outside vendors
Chase says some customer names and emails may be affected. A ransomware gang called n0n has claimed the attack and leaked data it says came from Chase.
Agents2 min
Signal blocks screenshots to keep phone AI agents out of private chats
Signal says built-in agents like Gemini and Siri AI threaten its privacy. The move follows a Reuters report that Meta's Muse agent read Signal messages.
Cybersecurity2 min
F5 issues emergency fixes for critical BIG-IP remote code execution flaws
The bugs can crash BIG-IP's traffic engine and may let unauthenticated attackers run code. F5 lists no mitigations besides the hotfixes.