Organizations with internet-facing Microsoft IIS web servers, or long-term AWS access keys sitting in application files, have a new reason to audit both. Attackers are using an open-source AI pentesting tool called AntiFox to find and use those secrets.
Security firm Arctic Wolf disclosed the findings on GitHub. It said unnamed threat actors hosted their own AntiFox instances on Alibaba Cloud infrastructure and used the tool's automation to analyze compromised systems, search for passwords, and reach cloud resources. AntiFox was built by a Chinese developer. Penetration testing, the practice of probing systems for weaknesses, is normally defensive work, and tools built for it are not malicious by design.
In some attacks, the intruders first compromised IIS servers, though the researchers could not determine how. They then installed web shells, small programs that let an attacker run commands on a server remotely, and combed application files and configuration settings for database logins and other secrets. Arctic Wolf also saw attempts to use those credentials against internal production databases. In other cases, attackers used valid AWS credentials obtained elsewhere to map accounts, permissions, databases, and storage servers.
Amazon said "AWS services were not compromised and operated as designed." It pointed customers to least-privilege permissions, IMDSv2 on EC2 instances, keeping long-term keys out of application configurations, and Amazon GuardDuty.
That framing fits the facts. Nothing described here is a new hole in AWS. The entry point is an ordinary server compromise or a leaked key, and the AI tool's role is the automated searching that follows. It is the same pattern as Cobalt Strike and ConnectWise ScreenConnect, legitimate tools that attackers adopt.
Arctic Wolf advises watching web servers for random new files or unapproved operating-system commands, terminating unfamiliar web shells, and rotating every exposed password, API key, and cloud credential. In AWS, it flags rapid requests across multiple services and attempts to reach stored credentials from unfamiliar infrastructure.
AntiFox is a fork of ARTEX 0.1.1, another Chinese-developed AI pentesting framework. Censys found 1,465 hosts running exposed ARTEX consoles, and about 10% also ran at least one other tool commonly used in cyberattacks.