Security News

AI Agents Found and Exploited a Zero-Day Without Human Help in a Suspected Chinese Campaign

The Register · 21 Nov 2026
Key Takeaway Assume automated attackers will persistently probe every internet-facing service, so patch quickly, remove unneeded exposure and enforce strong, unique credentials with multi-factor authentication.

Suspected Chinese spies used AI agents to find a previously unknown flaw in an open source application and then use it to break into university networks, according to Amazon CISO CJ Moses. The agents worked autonomously and gained full control of a server.

The attacker used Cairn, an open source penetration-testing framework that Moses says is increasingly popular with cybercriminals for multi-stage intrusions with no human assistance. Amazon's Threat Intelligence Team found a publicly hosted Cairn instance on September 10 that gave visibility into a 48-hour campaign beginning on September 8, targeting universities in East Asia outside Mainland China. Prompts were written in Simplified Chinese and script timestamps used China Standard Time.

The agents did not succeed straight away. Over about 16 hours and 148 sequential actions, they tried to compromise a university's Redis server, then tried a Log4Shell injection, a server-side request forgery attack against a Qdrant instance, and credential sprays against PostgreSQL, SSH and other services. A separate operation ran for more than 6 hours against a GPU server, using a list of 80 real user names gathered from public websites and academic papers. Neither succeeded, but Moses said they show the agents adapted for nearly a full day without human direction.

The agents then found a path traversal zero-day in MONAI Label, an open source medical imaging AI application. They mapped its API, found an upload endpoint with a filename parameter, and went from mapping to writing arbitrary files in about 11 minutes. They later ran a Python reverse shell as root. A human operator then supplied a proxy into a second university network, where a server ran the same vulnerable code. Amazon notified the maintainers, who released a fix and advisory. Gambit Security separately reported a Chinese operator using Cairn in at least 105 attacks, compromising at least 27 companies to varying degrees.

Why it matters

Defences that count on attackers giving up after failed attempts or lacking the time to study obscure software are weakened when agents can keep adapting for a full day and move from API mapping to file writes in minutes. Anyone running open source AI or data tools on internet-facing servers is exposed, especially where the same vulnerable code runs on several networks. It pushes teams to treat unusual, sustained failed login and exploit activity as a serious warning rather than background noise.

What to do

  • Apply the MONAI Label fix and read the maintainers' security advisory if you run the application.
  • Review logs for long runs of failed logins against PostgreSQL, SSH and similar services.
  • Check for unexpected file uploads or writes on servers hosting MONAI Label.
  • Restrict public exposure of Redis, Qdrant and database services to only what is needed.
AI agents zero-day Cairn MONAI Label China Amazon
Running AI agents? Eight controls for securing them, mapped to the ISM ->

Summarised by CISO AI from The Register, written with Claude Sonnet 5.5. We link back to every original so you can read it yourself.