Security

AI Agents Found and Exploited a Zero-Day on Their Own to Take Over a University Server, Amazon Says

Leave a comment

Amazon CISO CJ Moses says suspected China-linked attackers used Cairn, an open-source AI penetration-testing framework, to run agents that worked for nearly a day with no human direction, The Register reports. The campaign targeted universities in East Asia over 48 hours starting Sept. 8. Amazon found it through a publicly exposed Cairn instance; the prompts were in Simplified Chinese.

The agents first spent about 16 hours and 148 steps trying to break into a Redis server, then tried Log4Shell, an SSRF against a Qdrant instance and credential sprays. All failed. Then they found a zero-day path traversal in MONAI Label, an open-source medical imaging AI app, and got control of a server. Amazon told the maintainers, who shipped a fix and an advisory.

My take

“Neither operation succeeded, yet… the agents kept adapting.” That’s the part that matters. Attackers no longer get tired. So the basics matter more than ever: don’t expose Redis, vector databases or AI tools to the internet, patch your apps, and assume anything public will get tested around the clock.

God Bless, and Use the Source!

References & Further Reading

Subscribe (RSS) Follow Security via RSS

Leave a note

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.