AI agents find zero-day in 11 minutes and take root on medical-imaging server

Suspected Chinese hackers gave an AI agent one prompt to get shell access, and it found a brand-new vulnerability in 11 minutes.

Tools · Source: BleepingComputer

What happened

Suspected Chinese-linked hackers used an autonomous AI tool called Cairn to attack educational institutions in East Asia. Cairn is an open-source penetration testing framework that coordinates multiple AI agents. The attackers gave the AI a single plain instruction to obtain shell access. The AI agents operated independently to map the target application and find a vulnerability.

The primary target was MONAI Label, an open-source medical imaging server. In just 11 minutes, the AI agents discovered a previously unknown path-traversal flaw. They used this zero-day vulnerability to write files outside the approved storage folder. The software runs with no authentication enabled by default, allowing any client reaching the server to execute write commands.

In a later session, the agents escalated the attack by planting a Python script that runs automatically at startup. This gave the attackers complete root access to a university server. Amazon Threat Intelligence uncovered the operation because the hackers left their own Cairn control panel exposed to the public internet without a password. The exposed panel revealed the task settings, prompts, and run history of the agents.

Key facts

Why it matters

Autonomous AI agents cross the line from assisting hackers to executing attacks end-to-end. You no longer need human intuition to find a zero-day. You just need a target and a prompt. Security teams are now racing against machines that map APIs and test exploits at superhuman speeds. This is reportedly the first documented case of AI agents finding a brand-new security hole entirely on their own. The barrier to entry for devastating cyberattacks just dropped to zero.

The defense landscape shifts entirely for anyone building or hosting software. Manual investigations cannot keep up with an AI that tests thousands of attack vectors in minutes. If your application runs with default settings and no authentication, an AI agent will find it and break it before your monitoring tools even trigger an alert. Amazon explicitly warned that organizations must rely more on automated defenses. Human reaction times are no longer sufficient to stop an autonomous breach.

For builders

Turn on authentication by default

MONAI Label ships with authentication turned off. If you build open-source tools, force users to set a password on installation. Builders who rely on default open ports will lose their servers to automated agents.

Restrict application write permissions

The AI succeeded because the server could write to executable locations. Run your applications under limited service accounts, not root. You pay the price when a path-traversal flaw lets an AI drop a reverse shell into your startup scripts.

Audit automated defense systems

Amazon warns that AI agents find vulnerabilities faster than humans can react. You must deploy automated defenses to block rapid API mapping. Companies relying on manual log reviews will lose data before the security team wakes up.

My take

I always tell founders to assume their code is broken, but now the machines are proving it. We are officially in the era where AI agents find and exploit zero-days while we sleep. Build automated defenses today, or watch an autonomous script steal your root access.

Original reporting: BleepingComputer. This is my rewrite and opinion.

More AI news for builders