AI Aulis

AI NEWS IN ENGLISH · 21 November 2026

The most important AI news every morning. AI writes and checks the stories against the original sources – check important facts in the sources.

← Issue 21 November 2026Paper

Amazon: AI agents went from mapping an API to exploiting a flaw in about 11 minutes

AI-generated contentWritten by AI

aws.amazon.com, news.fmisec.com, theregister.com, bleepingcomputer.com, bankinfosecurity.com, mallory.ai · 7 sources · Sources published: 20–21 November 2026

Amazon says AI agents used by an attacker linked to China found an unknown flaw in a medical imaging application and gained root access to a server. A fix was released on 18 November.

Amazon Threat Intelligence wrote on the AWS blog on 20 November about a campaign in which an attacker gave AI agents a list of IP addresses and the goal ‘get a shell’. The agents ran under Cairn, an open-source penetration-testing tool that coordinates several agents towards a given goal. Amazon was able to examine the operation’s history because the attacker’s Cairn server was exposed to the internet without authentication. Anyone could read the prompts, settings and the agents’ output.

The campaign targeted universities in East Asia outside mainland China on 8–10 September. Amazon assesses with moderate confidence that the attacker is linked to China. It bases this on prompts written in Simplified Chinese, timestamps in China’s time zone and the choice of targets.

The agents mapped the API of a medical image analysis application. They then found that adding a directory path to a file name in the upload API let them write a file anywhere on the server. Going from mapping to writing files took about 11 minutes. In a later session, the agents wrote a reverse shell payload to sitecustomize.py, a file Python loads automatically at startup. This gave them root access to the server. The Register and BleepingComputer report that the application is MONAI Label. Amazon assesses with high confidence that the flaw is a zero-day vulnerability, because it found no fix, disclosure or mention of it in vulnerability databases.

Two other operations that Amazon described failed. The agents tried to break into a Redis database server for about 16 hours, taking 148 actions in that time. They tried more than 30,000 credential combinations against remote login on a university GPU server and other university systems without finding one that worked.

The agents did not move into a second university network on their own. The attacker gave them a proxy into that network. The same vulnerable software was running behind it. Amazon concluded that the attacker had compromised the server through the same flaw. A Recorded Future News article, republished by fmisec.com, presents this compromise as fact.

Amazon said an agent that finds and exploits a zero-day in minutes finishes its work faster than many security operations centres can handle an alert manually. The company therefore recommends reliable attack detection that triggers a response automatically. BleepingComputer reported that MONAI Label’s maintainers released a fix on 18 November. The threat intelligence service Mallory urges users of the application to install the update and check their publicly accessible installations for unauthorised file writes and remote connections.

Report an error in this story