AWS Details Cairn AI Agents’ Zero-Day Exploit by Chinese-Nexus Actor

Summary
AWS says a Chinese-nexus actor used Cairn AI agents to target East Asian universities; agents found an unpatched path-traversal flaw in medical-imaging software and gained root access. AWS says maintainers fixed the flaw and issued an advisory.
Key points
- AWS identified a publicly accessible Cairn instance whose operational history exposed the actor’s prompts, configurations, and AI output.
- The agents targeted university systems for about 48 hours, including extended, unsuccessful attempts against Redis and SSH services.
- Against an open-source medical-imaging AI application, agents found a path-traversal flaw and wrote files to the server in about 11 minutes.
- In a later session, the agents used the flaw to place a Python reverse-shell payload in sitecustomize.py and trigger it, gaining a root shell.
- AWS assesses with high confidence that the flaw was a zero-day at discovery; the project’s maintainers were notified, released a fix, and issued a GitHub Security Advisory.
- AWS assesses with moderate confidence that the actor was Chinese-nexus, citing Simplified Chinese prompts, China Standard Time timestamps, and targeting choices.
- AWS recommends high-confidence detections that trigger automated response, with human oversight, because agent-driven activity can move faster than manual triage.
Article Details
- Attack Vectors
- Agents exploited an unauthenticated file-upload endpoint vulnerable to path traversal in an open-source medical imaging AI application, enabling arbitrary file writes.
- Agents wrote a Python reverse-shell payload to sitecustomize.py and triggered a training API endpoint; the resulting process imported the payload and opened a root-level reverse shell.
- Agents attempted to abuse Redis replication, escape its scripting sandbox, modify configuration files, and corrupt memory; these attempts failed.
- Agents used SSRF in a co-located Qdrant instance to map an internal Docker network, then attempted credential sprays against PostgreSQL, SSH, and other services; the attempts failed.
- Agents tested more than 30,000 credential combinations against SSH and other university systems without finding a working login.
- The actor provided a proxy that tunneled into a second university network and directed agents to target other machines there; the article infers the actor had used the zero-day to compromise the proxy-connected server.
- Defensive Notes
- Prioritize high-confidence detections that can trigger automated response, with humans overseeing outcomes rather than reviewing every individual action.
- AWS describes using global honeypots and network traffic analysis to identify scanning for vulnerable systems and help restrict that activity at the network level.
Indicators of compromise
| Type | Indicator | Context |
|---|---|---|
| IPV4 | 142[.]4[.]107[.]34 | Listed in the article's indicators of compromise for records dated September 8–10, 2026; its specific role is not stated. |
| IPV4 | 142[.]4[.]98[.]145 | Listed in the article's indicators of compromise for records dated September 8–10, 2026; its specific role is not stated. |
MITRE ATT&CK
T1046 · Network Service DiscoveryAfter exploiting SSRF in a co-located Qdrant instance, the agents mapped the internal Docker network topology and probed services including PostgreSQL and SSH.T1059.006 · PythonThe agents wrote a Python reverse-shell payload to sitecustomize.py; a newly spawned Python process imported it and established a root-level reverse shell.T1090 · ProxyThe actor supplied the agents with a proxy that tunneled into a second university network for targeting other machines.T1110.003 · Password SprayingThe agents tested credential combinations against SSH, PostgreSQL, and other services, including more than 30,000 combinations against a university GPU server; the login attempts failed.T1190 · Exploit Public-Facing ApplicationThe agents exploited a path traversal vulnerability in the target medical imaging application's upload endpoint to write files on the server.
Vendors
Products
DockerWhen those attempts failed, the agents mapped the internal Docker network topology through a server-side request forgery (SSRF) vulnerability in a co-located Qdrant instance and launched credential sprays againstPostgreSQLforgery (SSRF) vulnerability in a co-located Qdrant instance and launched credential sprays against PostgreSQL, SSH, and several other services.QdrantDocker network topology through a server-side request forgery (SSRF) vulnerability in a co-located Qdrant instance and launched credential sprays against PostgreSQL, SSH, and several other services.RedisThe most extensive operation targeted a Redis (remote dictionary server) instance at one university that was configured to allow access without authentication.
