AI agents exploited a MONAI Label zero-day before a patch was released
Amazon says autonomous AI agents, working with minimal human involvement, discovered and exploited a path-traversal zero-day in MONAI Label during a campaign against universities in East Asia outside mainland China. The flaw allowed arbitrary file writing, and Amazon said the agents reached it in about 11 minutes.
Key takeaways · 4
- 01
Amazon said the agents found and exploited the MONAI Label flaw in about 11 minutes.
- 02
The flaw affected datastore upload endpoints and allowed files to be written outside the intended storage directory.
- 03
Authentication is disabled by default, and an authenticated user with the ANNOTATOR role can still exploit the flaw.
- 04
The fix was merged into the main branch, but the next MONAI Label release is expected to include it.
What the agents did
Amazon Threat Intelligence said it investigated a Chinese-nexus threat actor using autonomous AI agents with minimal human involvement.[1] Amazon said the campaign ran for about 48 hours from September 8 and targeted universities in East Asia outside mainland China.[2] Amazon assessed with moderate confidence that the activity had a China nexus.[2] The agents discovered and exploited the MONAI Label flaw in about 11 minutes, according to Amazon.[2] Cairn, the framework used to coordinate the agents, is an open-source framework for security testing.[2]
The flaw and its impact
MONAI Label is an open-source medical-imaging AI application.[2] The vulnerability was a path traversal in datastore upload endpoints that allowed arbitrary file writing.[3] Amazon said the flaw was used to write files outside the intended storage directory.[2] In a later session, the agents used the same flaw to run malicious code and gain full control of a server.[2] Amazon assessed with high confidence that the vulnerability was a zero-day because it found no prior fix, issue, advisory or vulnerability-database mention.[2] That assessment describes what Amazon found; it does not establish that every MONAI Label installation was compromised.
Patch and exposure status
The fix was merged into MONAI Label’s main branch on October 19, but no patched release was available in the reporting.[2] The newest PyPI release listed was version 0.8.5, uploaded on November 25, 2024, and the advisory lists 0.8.5 and earlier without a patched version.[2] The fix is due in the next MONAI Label release.[2] Authentication is disabled by default, so anyone who can reach the server can exploit the vulnerability; even with authentication enabled, users assigned the ANNOTATOR role can exploit it.[2] The advisory recommends enabling authentication and limiting ANNOTATOR access to trusted users.[2]
A wider AI security picture
The MONAI Label case is one example of AI being used in security-related activity, but the cited reporting also describes AI-assisted vulnerability discovery and AI-supported targeting. Nature reported that AI-assisted researchers had found security flaws in Java and PHP software.[4] It also reported that hackers used AI tools to target supermarket fridges and freezers in Europe.[4] Separately, Nature said disruptions involving computer systems run by Microsoft, US air traffic control, JPMorgan Chase and the US National Security Agency had renewed concerns about critical digital infrastructure resilience.[4]
The report gives security and engineering teams a concrete reason to check whether MONAI Label is reachable and which users have the ANNOTATOR role. A merged fix is not the same as a patched release, so teams should verify their installed version and follow the project’s release and advisory updates.
Why it matters
AI training for your team
kju.ai is built for teams: daily sessions, team analytics, and rollout support for your organization.
How this developed
22 November 2026
Nature reported that AI-assisted researchers had found security flaws in Java and PHP software, and that hackers had used AI tools to target supermarket fridges and freezers in Europe.
21 November 2026
MONAI Label’s advisory identifies versions 0.8.5 and earlier as affected, while follow-up reporting says the fix was merged but the newest PyPI release was still 0.8.5.
21 November 2026
Amazon says Cairn agents exploited a MONAI Label zero-day
Sources
- China attackers deployed AI agents that autonomously found and exploited a 0-day to take over a university servertheregister.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
- Hackers use autonomous AI tool Cairn to find zero-day, take over serverbleepingcomputer.com
- China attackers deployed AI agents that autonomously found and exploited a 0-day to take over a university servertheregister.com
- Chinese AI Agents Exploit Zero-Day in Medical Software - SecurityWeeksecurityweek.com
- Path traversal in datastore upload endpoints allows arbitrary file write · Advisory · Project-MONAI/MONAILabel · GitHubgithub.com
- Amazon says attackers' AI agents found and exploited a zero-day in 11 minutes, but no patched release is available | VentureBeatventurebeat.com
- Chinese-nexus actor’s Cairn AI agents found a zero-day on their own | AWS Security Blogaws.amazon.com
- AI is turbocharging cyberattacks — can computer scientists catch up? | Naturenature.com
