Most requests were denied. AntiFox-linked intrusions still reached production cloud secrets.

A flood of access-denied responses did not mean the intrusions had failed. Arctic Wolf found infrastructure associated with the AI pentesting platform AntiFox using valid AWS identities to explore cloud resources, with successful access to a production secret among the observed activity. In one intrusion, multiple identities generated hundreds of requests across cloud services in roughly half an hour.

The investigation connected compromised websites with subsequent cloud activity. Researchers saw a remote-control script placed on an IIS web server and application configuration and database connection details being read. Other activity involved valid identities reading, copying and deleting S3 storage objects. The concern extends beyond one exposed file: credentials kept inside a web application can provide a route into the cloud systems supporting a business.

AntiFox is an open-source, Chinese-language AI penetration-testing platform, a category intended to help identify security weaknesses. This report links infrastructure hosting its interface to malicious actions. Its practical warning is that successful operations can disappear into a much larger pile of rejected requests. Defenders therefore need to correlate unusual web-server behavior, unfamiliar sources of cloud access and successful secret retrieval, rather than treating a high failure rate as proof of safety.

There are limits to what the evidence establishes. Arctic Wolf did not identify the initial IIS vulnerability, and the report does not demonstrate that AI independently chose every action. Amazon told BleepingComputer that AWS services were not compromised: the attackers used credentials taken from customers' compromised applications. The finding concerns the combination of existing tools, stolen credentials and cloud permissions, not an AI breakthrough against AWS itself.

Source times: Arctic Wolf Labs report first repository commit, Nov 11, 2026, 00:41:08 Beijing / Nov 10, 2026, 11:41:08 AM EST; BleepingComputer original publication, Nov 11, 2026, 23:12:59 Beijing / Nov 11, 2026, 10:12:59 AM EST, updated Nov 12, 2026, 13:12:42 Beijing / Nov 12, 2026, 12:12:42 AM EST.