Universities in East Asia agentic AI attack — Sep 2026
Amazon uncovered a campaign where autonomous AI agents used the Cairn framework to discover and exploit a zero-day vulnerability in a medical imaging.
- Incident date
- Sep 8, 2026
- Source report date
- Nov 20, 2026
- Target
- Universities in East Asia
- Agent type
- Other AI agent
- Agent role
- Used by the attacker
Autonomous AI Exploitation Campaign
In September 2026, Amazon’s threat intelligence team identified a campaign where Chinese-nexus actors utilized the Cairn autonomous AI penetration-testing framework to target universities in East Asia. The agents operated with minimal human intervention, planning and executing complex attack sequences to gain root access to target systems.
What happened
Beginning on September 8, 2026, the threat actors deployed Cairn to target university IP addresses. Amazon discovered the operation after finding a publicly accessible, unauthenticated server hosting the Cairn instance, which exposed project configurations, prompts in Simplified Chinese, and operational logs. The agents were given a single instruction: obtain shell access. From that point, the AI autonomously planned and adjusted its tactics without further human guidance.
During the 48-hour campaign, the agents demonstrated persistent, multi-stage attack capabilities. In one instance, they spent 16 hours attempting 148 different actions against a Redis server, including memory manipulation and exploiting Log4Shell, before pivoting to an SSRF vulnerability in a Qdrant server. In another attempt, the agents scraped over 80 usernames from university websites and performed more than 30,000 credential stuffing attempts against SSH and VPN services.
The agents successfully identified and exploited a zero-day vulnerability in an open-source medical imaging application. Within 11 minutes of mapping the application’s API, the agents identified a path traversal vulnerability in an upload parameter. They subsequently achieved a root shell by writing a Python reverse shell to sitecustomize.py and triggering a training endpoint. Amazon assessed this as a genuine zero-day, as no prior documentation or patches existed for the vulnerability. Once the agents secured the initial foothold, the human operator provided a proxy tunnel to a second university network, allowing the agents to leverage the same zero-day to compromise additional infrastructure. Amazon notified the project maintainers, who subsequently released a fix. The report highlights the speed of these autonomous operations, noting that AI agents can identify and exploit vulnerabilities faster than traditional security teams can respond to individual alerts.
Evidence in the reporting
- Incident evidence
- autonomous agents found a zero-day and gained root access
- Agent involvement
- The agents planned, executed, and adjusted on their own