The short answer
- Amazon says AI agents linked to a Chinese-nexus actor found and exploited a zero-day in MONAI Label, an open-source medical imaging app, during attacks on universities in East Asia.
- Most of the campaign failed, including attempts against a Redis server and a credential-spray effort, so the reports do not show a broad AI breakthrough.
- Google's GTIG reports that attackers used an AI framework called exploit_agent to turn security patches into exploits with no human in the loop.
- Attribution is unconfirmed, and the sources reviewed do not include a government advisory specifically about Cairn.
A zero-day is a flaw that, in the case discussed here, had no advisory or patch when it was found. Amazon's report of November 20, 2026 says AI agents found and exploited one during attacks on universities, while a Google report of November 16 separately describes attackers using AI to turn patches into exploits. Amazon's findings are its own assessment, and attribution remains unconfirmed.
What a zero-day vulnerability is
The flaw versus the exploit
As a general explanation rather than a sourced definition, a vulnerability is a weakness in software that can be abused, and an exploit is the method or code that takes advantage of it.
Why it's called 'zero-day'
The sources reviewed for this article do not explain where the term comes from. Readers who need a formal definition should check official glossaries, such as those from NIST or CISA, because this article could not verify one.
Why zero-days are hard to defend against
Amazon said the path traversal flaw had no advisory and no patch when it was found, so there was no fix to install at that point. As a general explanation, updates can only deliver fixes that exist.
What the Amazon report found
On November 20, 2026, Amazon's CISO CJ Moses published a report on the AWS Security Blog. It says Amazon Threat Intelligence observed a Chinese-nexus threat actor using autonomous AI agents with minimal human involvement. Amazon says the operator's instruction was simply "obtain shell access."
The campaign
The tool at the centre is Cairn, an open-source autonomous AI penetration-testing framework that Amazon says was built as a research tool for defenders. Its agents plan and adjust steps without the operator choosing each one, which Amazon says makes it attractive to attackers. Amazon first identified the operator's publicly exposed Cairn instance on September 10, 2026.
The campaign targeted IP addresses tied to universities in East Asia outside mainland China, over about 48 hours starting September 8, 2026. AWS declined to name the universities or the country, according to The Record.
The zero-day
According to The Record, the only successful attempt was the agents finding and exploiting a flaw in MONAI Label, an open-source medical imaging AI application. The Register reports that the maintainers released a fix and a security advisory, but the sources reviewed do not give the fix date.
Amazon assesses with high confidence that the flaw was a zero-day, based on the absence of any fix, issue, advisory or prior mention in vulnerability databases. That is Amazon's assessment, not a formal confirmation from the vendor or a government agency.
What failed
Most operations did not succeed. The most extensive hit a Redis instance at one university, where over roughly 16 hours and 148 actions the agents tried known techniques without success. A credential-spray attack against a university GPU server, using 80 real usernames, also failed, according to The Register.
The Register paraphrases a CISO saying the agents kept adapting for nearly a full day without human direction. That refers to the failed operations, not to time spent exploiting the zero-day.
The second university
After the zero-day, the human operator gave the agents a proxy into a second university network with a server running the same vulnerable code, according to The Register.
Why AI changes the picture
Google's findings
Google Threat Intelligence Group's quarterly report, "Agentic Threat Activity Crosses a Significant Threshold," was published on November 16, 2026, and SiliconANGLE covered it. GTIG says attackers used an AI framework named exploit_agent, with no human in the loop, to turn security patches into exploits.
GTIG also thinks attackers may find it more efficient to target n-day flaws, which are already disclosed and have fixes available, because the source material for patched flaws is finite. The same report says Cairn was observed in an intrusion against the U.S. aerospace sector.
Writer's conclusion, not a GTIG recommendation: because GTIG's view points toward n-day targets, checking for disclosed flaws that already have fixes is a sensible priority.
Speed and persistence
Amazon argues that agentic attacks are already happening and that defenders should prepare for their speed and persistence. The Record reports that a Gambit Security researcher considers Cairn more significant than the Artex tool because it is general-purpose and less predictable.
Why it's trending
Coverage clustered on November 16 and 20, 2026. The headline claim is that an AI agent found a zero-day without a human choosing each step, which is why readers are asking whether this is new and whether they are at risk. The Record also reports that the FBI published an advisory about a separate AI hacking tool, Artex, about a week earlier.
What happens next
Patching and advisories
According to The Register and The Record, the MONAI Label maintainers released a fix and a security advisory, but the sources reviewed do not give a date. As of November 21, 2026, no government advisory specifically about Cairn appears in the sources reviewed.
Attribution and government response
Attribution remains unconfirmed. AWS describes the actor as Chinese-nexus, The Register calls the attackers suspected, and The Record uses the word allegedly. AWS worked with a government partner to notify affected organisations, and no count of those organisations has been published.
What readers can do
Readers who run MONAI Label should check the project's security advisory and release notes rather than relying on news reports. For campaign updates, watch for official statements from Amazon and GTIG.
Readers in the United States, the United Kingdom, Canada and Australia should note that the reported victims were universities in East Asia. The sources reviewed are US-based, no UK, Canadian or Australian government response was found, and local guidance was not checked.
Questions people ask
Was the MONAI Label flaw confirmed as a zero-day?
Amazon assesses with high confidence that it was a zero-day, based on the absence of any fix, issue, advisory or prior mention in vulnerability databases. That is Amazon's own assessment, not a formal confirmation from the vendor or a government agency.
What is Cairn?
Cairn is an open-source autonomous AI penetration-testing framework, which Amazon says was built as a research tool for defenders. Its agents plan and adjust steps without the operator choosing each one. Amazon says that ease of use makes it attractive to attackers.
Did the attack succeed?
Most attempts failed. According to The Record, the only successful attempt was the agents finding and exploiting the MONAI Label flaw. The Redis and credential-spray operations did not succeed, according to The Register.
Who is behind the campaign?
Attribution is unconfirmed. AWS calls the actor Chinese-nexus, The Register says suspected, and The Record says allegedly. AWS declined to name the universities or the country involved.
Sources
- Chinese-nexus actor's Cairn AI agents found a zero-day on their own — AWS Security Blogaws.amazon.com
- China attackers deployed AI agents that autonomously found and exploited a 0-day to take over a university server — The Registertheregister.com
- Amazon says Chinese hackers used Cairn AI tool to discover apparent zero-day during attack on university — The Record from Recorded Future Newstherecord.media
- Google says an AI framework turned security patches into exploits with no human in the loop — SiliconANGLEsiliconangle.com



