Home / Blog / Amazon: China-Linked Hackers’ AI Agents Exploited Medical…
Tech News

Amazon: China-Linked Hackers’ AI Agents Exploited Medical Software Zero-Day

Amazon revealed China-linked hackers deployed autonomous Cairn AI agents to exploit a critical zero-day bug in MONAI Label across East Asian universities.

In short

Healthcare IT teams and academic network administrators · 11 minutes from API mapping to arbitrary file write · Fix scheduled for next MONAI Label release · Restrict network access to MONAI Label installations

By Dillip Chowdary • Nov 21, 2026 • Source: SecurityWeek

Amazon: China-Linked Hackers’ AI Agents Exploited Medical Software Zero-Day
Image: SecurityWeek

At a glance

Who it is for
Healthcare IT teams and academic network administrators
The number
11 minutes from API mapping to arbitrary file write
Available
Fix scheduled for next MONAI Label release
What to do
Restrict network access to MONAI Label installations

Amazon revealed that a China-linked threat actor used autonomous Cairn AI agents to discover and exploit an unpatched zero-day vulnerability in MONAI Label version 0.8.5 and earlier during a 48-hour attack window in September, according to SecurityWeek's report.

This report details how autonomous penetration-testing frameworks weaponize path traversal flaws against medical imaging systems and unauthenticated network endpoints. Healthcare IT administrators, research institute security teams, and infrastructure engineers can review the attack sequence and inspect exposed deployments before the upcoming software release.

01Exposed framework logs and campaign discovery

Amazon Threat Intelligence uncovered the intrusion activity after locating an exposed deployment of Cairn, an open source penetration testing framework powered by artificial intelligence agents. The system operated without basic authentication safeguards, leaving operator prompt logs, agent commands, and tool responses accessible to observers. Reviewing the unauthenticated records exposed a continuous 48-hour offensive operation focused on academic targets.

The threat group directed the automated tools against IP address blocks belonging to universities located across East Asia outside mainland China. Operators loaded target inventories into the framework with instructions commanding the software to obtain root-level shell access. The agents then initiated reconnaissance without manual intervention, probing network services and profiling target applications.

Amazon assessed the zero-day status with high confidence because the target open source repository had no prior issue tickets, vulnerability database entries, or security advisories documenting the flaw. Attributing the activity with moderate confidence to an unidentified Chinese-nexus threat actor, investigators cited prompt phrasing, script time zone indicators, and the strategic profile of the targeted institutions.

02Eleven minutes to arbitrary file write

Upon identifying MONAI Label on the target host, the Cairn framework mapped the application programming interface to find input paths. The agents identified an unprotected file upload mechanism and isolated a path traversal vulnerability within the service. The software leveraged this weakness to bypass folder boundaries and place arbitrary payloads into the target filesystem.

The autonomous agent completed the entire discovery and initial compromise sequence in roughly 11 minutes. It transitioned from initial API exploration directly to planting unauthorized files on the remote server without human operator guidance. The speed of the process bypassed traditional manual penetration testing cycles.

MONAI Label functions as an open source utility designed for AI-assisted annotation of medical images, leaving research infrastructure open to code execution. Because authentication is disabled by default across deployments, any network entity capable of establishing a connection to the server can interact with the API and trigger the traversal condition.

Advertisement

Tech Pulse Daily

Get tomorrow's pulse first

Join engineers who read Tech Pulse before stand-up. Free, weekday mornings.

03Python interpreter hijacking and root shell

  1. Map application endpoints automated agents scan network interfaces and document accessible API functions.
  2. Locate traversal flaw agents submit test payloads to the file upload routine to identify directory traversal opportunities.
  3. Plant reverse shell the system writes a Python reverse shell payload into a startup script evaluated whenever Python initializes.
  4. Trigger interpreter execution the application starts a Python process, executing the dropped script with elevated permissions.
  5. Establish root control the reverse shell calls out to the actor infrastructure, granting persistent administrative access.

Note: MONAI Label versions 0.8.5 and earlier contain the critical path traversal vulnerability, which allows remote code execution because network authentication is disabled by default.

04Lateral movement and network tunneling

Following initial host compromise, the human operator provided the agent instance with a network proxy configured to tunnel directly into a second university system. Instructions directed the toolset to pivot through the newly established foothold and seek additional target nodes across the secondary network. The agents began scanning adjacent internal IP ranges to identify reachable services.

Network mapping confirmed the secondary server ran the identical vulnerable MONAI Label build. Investigators inferred the threat actor had previously leveraged the zero-day flaw to penetrate that server and set up the tunnel connection used for lateral scanning. The automated toolset systematically validated the software stack across target institutional boundaries.

The attack demonstrate autonomous tools executing end-to-end intrusion playbooks, moving from external scanning to lateral pivoting. By chaining exposed medical annotation tools with internal network proxies, the actor expanded access across regional academic infrastructure without manual command execution.

05Project advisory and deployment status

The development team maintaining MONAI Label published a security advisory confirming the critical severity of the flaw across all installations up to version 0.8.5. A software patch has been engineered to eliminate the path traversal weakness and enforce upload boundary controls. The remediation will be distributed to public repositories in the next project release.

Administrators running MONAI Label in research laboratories and clinical facilities must isolate existing instances from untrusted local networks and the public internet. Enabling network authentication, implementing strict access control lists, and monitoring file write activities around Python startup libraries mitigate exposure until the patched build releases.

Security teams monitoring academic infrastructure should also review external perimeter exposures for testing frameworks and medical AI pipelines. Audit logs, active tunnels, and outbound connections from image annotation hosts should be analyzed for signs of automated exploitation attempts matching the September campaign.

Frequently asked questions

What vulnerability was exploited in the attack?

Attackers exploited a critical zero-day path traversal flaw in the file upload function of MONAI Label version 0.8.5 and earlier.

How did the autonomous agents achieve root access on the server?

The agents wrote a Python reverse shell into a file that executes whenever the Python interpreter starts, spawning an administrative shell.

Which organizations were targeted during the campaign?

The threat actor targeted IP addresses belonging to universities in East Asia located outside mainland China during a 48-hour campaign.

Sources

Dillip Chowdary

Author

Dillip Chowdary

Writes Tech Bytes coverage of AI, engineering, and the tools that actually ship. Editor of Tech Pulse Daily.

Related on Tech Bytes

Advertisement

5-min tech signal

Weekday briefing for engineers who skip the noise.

No spam · Unsubscribe anytime

Advertisement

✈️ CareerPilot

Your AI job-search copilot

Match your resume against live Ashby, Greenhouse & Lever openings — fit scores, job-specific resume optimization and email alerts.

Find matching jobs →

Free Tools

Browse all tools →