Disclosure: this page contains affiliate links. East Bay Cyber may earn a commission if you buy through them, at no extra cost to you. That never changes how we rank products. Editorial policy.
Amazon says Cairn agents found zero-day in university attack
TL;DR - Amazon says a Chinese-nexus threat actor used Cairn, an open-source autonomous AI penetration-testing framework, against university-associated IP addresses in East Asia. - The agents found an unpatched path traversal vulnerability in an open-source medical imaging AI application in about 11 minutes. - Amazon says the agents used the flaw to write files and establish a reverse shell running as root. - The affected project’s maintainers released a fix after AWS notified them. No CVE ID is included in Amazon’s advisory.
What happened#
Amazon Threat Intelligence says it identified a publicly hosted Cairn instance running without authentication on September 10, 2026. The exposed instance revealed project configurations, prompts, and AI output. The operator used Simplified Chinese prompts, and timestamps in executed scripts were configured to China Standard Time.
Amazon analyzed activity targeting IP addresses associated with universities in East Asia outside mainland China over an approximately 48-hour period beginning September 8, 2026. The actor gave the agents a plain-language objective to “obtain shell access.” Cairn coordinated multiple agents that planned, executed, and adjusted actions without the operator selecting each next step.
Some operations failed. Against an open Redis instance, the agents performed 148 sequential actions over roughly 16 hours, including attempts involving replication, scripting, configuration files, memory corruption, and Log4Shell injection. They also mapped an internal Docker network through a server-side request forgery vulnerability in a co-located Qdrant instance and attempted credential sprays against PostgreSQL, SSH, and other services. Amazon says none of those attempts succeeded.
In another operation, agents researched about 80 real usernames and tested more than 30,000 credential combinations against SSH and other university systems. Amazon says they did not find a working login.
The successful operation targeted an open-source medical imaging AI application. The agents identified the application framework, mapped its API endpoints, and tested a filename parameter on an upload endpoint for path traversal. In about 11 minutes, they progressed from API mapping to arbitrary file writes on the server.
During a later session, the agents wrote a Python reverse shell payload to sitecustomize.py, which Python automatically imports at interpreter startup. They then triggered a training API endpoint to start a new Python process. Amazon says the process established a reverse shell running as root.
Amazon says the flaw had no assigned security advisory and was unpatched when discovered. The company assessed with high confidence that it was a zero-day based on the absence of a corresponding fix, issue, advisory, or prior mention in vulnerability databases for the affected project. The project maintainers released a fix and a GitHub Security Advisory after AWS notified them.
AWS says the affected organizations were not AWS customers and that it worked with a government partner to notify them. The universities and the affected country were not named. AWS published these indicators for records dated September 8 to 10, 2026:
142[.]4.98.145142[.]4.107.34
Are you affected#
Organizations may be relevant to this incident if they run the affected open-source medical imaging AI application, particularly where its upload or training API is internet-accessible. Amazon’s advisory does not name the application or provide a version range, product version, or CVE ID.
AWS says the vulnerability was fixed after notification, but the available advisory does not identify the fixed version. Teams should therefore consult the affected project’s GitHub Security Advisory and current maintainer guidance before determining whether an installation is updated.
Organizations that do not run the affected application are not identified by AWS as affected by this particular vulnerability. However, the reported activity targeted university-associated systems and included Redis, Qdrant, PostgreSQL, SSH, VPN, network storage, campus portal, and administrative systems.
What is confirmed and what is reported#
| Claim | Source | Status |
|---|---|---|
| Amazon observed a publicly accessible Cairn instance running without authentication on September 10, 2026. | Amazon Threat Intelligence | Confirmed |
| Activity targeted IP addresses associated with universities in East Asia outside mainland China over an approximately 48-hour period beginning September 8, 2026. | Amazon Threat Intelligence | Confirmed |
| Cairn agents found a path traversal vulnerability in an open-source medical imaging AI application and used it to establish a reverse shell running as root. | Amazon Threat Intelligence | Confirmed |
| Amazon assessed with high confidence that the vulnerability was a zero-day when discovered. | Amazon Threat Intelligence | Confirmed |
| The affected project’s maintainers released a fix and a GitHub Security Advisory after AWS notification. | Amazon Threat Intelligence | Confirmed |
| The activity was attributed with moderate confidence to a Chinese-nexus threat actor. | Amazon Threat Intelligence | Confirmed |
| The attacker used the apparent zero-day to breach a second university. | The Record reports Amazon said the attacker used the flaw to breach a second university. | Reported |
| The universities and country involved were not named. | The Record reports that AWS declined to name them. | Reported |
What defenders should do now#
-
Identify deployments of the open-source medical imaging AI application, including instances operated by research groups, laboratories, and university departments.
-
Review the project’s GitHub Security Advisory and maintainer guidance, then apply the released fix. AWS has not published a version range or CVE ID in its advisory.
-
Determine whether upload and training API endpoints are reachable from the internet. Restrict access where it is not required and review authentication controls around those endpoints.
-
Search application, web server, operating system, and process logs for unexpected path traversal attempts, arbitrary file creation, writes involving
sitecustomize.py, unexpected Python process launches, and reverse shell activity. -
Investigate the indicators published by AWS,
142[.]4.98.145and142[.]4.107.34, in firewall, proxy, DNS, authentication, and endpoint telemetry for records dated September 8 to 10, 2026. -
Review exposed Redis, Qdrant, PostgreSQL, SSH, VPN, network storage, administrative, and campus portal services. Amazon’s investigation describes attempted access against these technologies and services.
-
Prioritize detections that can trigger automated containment or other rapid response, while retaining human oversight. Amazon says the agents progressed from application mapping to arbitrary file writes in about 11 minutes and continued unsuccessful activity for roughly 16 hours.
This advisory is informational and is updated as vendors publish details. Confirm versions and actions against the vendor’s own guidance for your environment.
Sources
- Primary notice: aws.amazon.com
- Primary notice: aws.amazon.com
- Research: wsj.com
- Research: crowdstrike.com
- Report: therecord.media