UK — The Register
China attackers deployed AI agents that autonomously found and exploited a 0-day to take over a university server
Security1 min read
Amazon says a China-linked hacking group used AI agents that found and broke into a flaw in a medical imaging app. The agents then took full control of a university server. The attack targeted universities in East Asia.
What happened
Amazon's chief information security officer, CJ Moses, said suspected Chinese spies used AI agents to attack universities. The agents worked on their own to find a zero-day, which is a flaw that the software maker did not know about yet. They used that flaw in MONAI Label, an open-source medical imaging app, to get full control of a server. Amazon told the app's maintainers, and they released a fix.
Key numbers and facts
- The campaign ran for 48 hours, starting on September 8.
- Amazon found the attacker's server on September 10.
- Before the zero-day, the agents spent about 16 hours and 148 actions trying to break into a Redis server. None of these worked.
- After the zero-day was found, the agents went from mapping the app to writing files on the server in about 11 minutes.
- The attackers' prompts were in Simplified Chinese, and the script timestamps used China Standard Time.
Background
Cairn is an open-source tool that tests computer security by trying to break in. Criminals have started using it too, because AI can run many steps of an attack with no human help. A separate report from Gambit Security also described a Chinese operator using Cairn in at least 105 attacks.
Why it matters
This shows AI agents can keep trying new ways to break in for almost a full day without a person guiding them. Attacks like this can happen faster and with less effort. Schools, hospitals, and other groups that run open-source software need to patch quickly and watch for unusual activity.
What to watch
- Whether more attacks use open-source AI testing tools like Cairn.
- Whether other users of MONAI Label install the fix.
- Whether more reports of AI-run attacks come from security firms.
Glossary
- Cairn
- An open-source tool for testing computer security by simulating attacks, which criminals have also started to use.
- MONAI Label
- An open-source app that uses AI to help label medical images.
- Redis
- A popular database tool that stores data in memory so it can be read quickly.