🚨Attackers Used AI Agents to Breach University Server Via MONAI Zero-Day in 48 Hours
Your AI agent could be someone else's pentester
TL;DR
Attackers used the open-source Cairn framework to run autonomous AI agents that exploited a zero-day in MONAI Label and took over a university server. The same framework has already hit at least 27 companies. If you expose ML tooling or agent frameworks, your threat model just changed.
AI agents just ran an entire intrusion campaign on their own. Amazon's threat team says attackers used the open-source Cairn penetration-testing framework to chain together multi-stage attacks against universities with almost no human involvement, starting September 8 and lasting 48 hours. If you run ML research infrastructure, this one is for you. The agents found a zero-day path traversal in MONAI Label, a popular medical imaging labeling tool, and used it to write arbitrary files and drop a root reverse shell. Anyone self-hosting MONAI Label or similar annotation platforms should treat their upload endpoints as hostile right now. The numbers are what make this uncomfortable. One agent spent about 16 hours and 148 sequential actions hammering another university's Redis server, trying replication abuse, sandbox escape, config writes, and memory corruption. None of it worked, but the persistence is the story. Separately, a credential spray against a GPU server used a list of 80 real usernames scraped from public pages and papers. The Simplified Chinese prompts and China Standard Time timestamps point to a Chinese operator, and Gambit Security previously tied the same framework to at least 105 attacks.

Key Points
Amazon's agents abused a zero-day path traversal in MONAI Label's upload endpoint, letting them arbitrarily write files on a university server starting September 8.
Keep reading — it's free
Enter your email to keep reading — plus the best of AI & tech, daily. Free, forever.
Already a member? Sign in
Comments
Be the first to comment