SMART WORLDJOURNALSubmit Manuscript
← All technology news

Artificial Intelligence

Autonomous AI Agents by Chinese-Nexus Actor Uncover Zero-Day Vulnerability Independently

Amazon Threat Intelligence reveals a Chinese-nexus adversary employing Cairn autonomous AI agents that discovered and exploited a zero-day flaw in an open-source medical AI application, underscoring the emerging speed and persistence of AI-driven cyberattacks.

NEWS

On September 10, 2026, Amazon Threat Intelligence identified a publicly accessible Cairn AI penetration-testing framework instance linked to a Chinese-nexus threat actor conducting autonomous hacking operations with minimal human oversight. This observer-accessible setup disclosed project prompts, configurations, and AI outputs in Simplified Chinese reflecting China Standard Time, enabling in-depth analysis of the campaign targeting East Asian university IP addresses over 48 hours.

Cairn is an open-source framework enabling multiple coordinated AI agents to achieve a specified goal by planning and executing actions autonomously. Intended for defensive research purposes, its ease of use also appeals to malicious actors. In this campaign, the operators instructed the agents to obtain shell access on selected targets, allowing the agents to independently select attack vectors and adapt tactics without further human input.

One operation targeted a Redis server at a university, attempting numerous known exploitation methods such as replication abuse, script sandbox escapes, and configuration tampering over 16 hours and 148 actions. Despite persistent credential spraying including against SSH and PostgreSQL, the attempts ultimately failed to gain access but demonstrated sustained autonomous attack persistence.

Conversely, the agents rapidly discovered a zero-day path traversal vulnerability in an unpatched open-source medical imaging AI application's upload filename parameter. Within 11 minutes, they achieved arbitrary file write capabilities and later leveraged that to deploy a Python reverse shell, gaining root privileges on the server. This zero-day had no prior advisory or fix, confirming its novel status.

After initial exploitation, the human operators enhanced the campaign by providing a proxy tunnel into a second university network running the same vulnerable software, suggesting lateral movement facilitated through the zero-day. Amazon Threat Intelligence attributes this activity with moderate confidence to a Chinese-nexus group based on language, timestamps, and strategic target profile.

Amazon notified the vulnerable project's maintainers, prompting a security advisory and patch release. Targeted universities, not AWS customers themselves, were informed through governmental channels. Additionally, the use of Cairn extends beyond espionage, having been identified in financially motivated operations targeting online retail as documented by Gambit Security, indicating broadening adoption of autonomous AI attack frameworks.

These developments pose challenges to security operations due to the swift discovery and exploitation capabilities of autonomous AI agents, compressing defense response windows. Maintaining effective security will require emphasis on high-confidence automated detection and response mechanisms to keep pace, supplemented by human oversight rather than manual action triage.

AWS incorporates active defense measures including global honeypots and network traffic analysis to detect such hostile scanning and limit adversary operations at the network level, providing protective layers integrated into AWS infrastructure at no cost to customers.

Source & further reading

Chinese-nexus actor’s Cairn AI agents found a zero-day on their own ↗

Source published 2026-11-20. Claims and limitations should be checked against the original announcement or research.