Hackers use AntiFox AI agents to steal AWS credentials
Attackers are hosting an open-source AI penetration testing tool on Alibaba Cloud to automate the discovery and theft of cloud secrets.
The short version
- AntiFox is an open-source agentic penetration testing platform that automates tasks like mapping services and probing for vulnerabilities.
- Attackers hosted the tool on Alibaba Cloud to target web services and harvest AWS secrets using valid identities.
- In one incident, hackers made hundreds of API requests in 30 minutes to probe AWS endpoints.
- The tool was originally created as a fork of ARTEX, another AI framework linked to cyberespionage and bank attacks.
- AntiFox remains on GitHub but lacks technical controls to prevent its use for illegal activities or live system attacks.
Why it matters
This highlights a shift toward 'agentic' security threats where AI can automate the tedious parts of a hack, like scanning for credentials. For businesses, it means attackers can move much faster from initial breach to full cloud environment takeover.
The rise of agentic hacking
Security researchers have identified a new pattern where hackers use self-managed instances of AntiFox to conduct automated attacks. AntiFox is an open-source platform designed for agentic penetration testing, which means it uses AI agents to perform complex security tasks. These agents can map out exposed services and probe various endpoints to find weaknesses without constant human input.
The tool was originally developed as a fork of ARTEX, a similar framework that gained notoriety for its use in cyberespionage and attacks on South Korean banks. While AntiFox features a Chinese-language interface, its capabilities are broad enough to facilitate various malicious activities. These include vulnerability exploitation, deploying webshells, and discovering configuration details.
The current trend shows a move toward automation where AI handles the heavy lifting of reconnaissance. Instead of a human manually checking every port or service, an AI agent can systematically scan an entire infrastructure. This allows attackers to cover more ground in a shorter amount of time than traditional manual methods would allow.
Infrastructure and methods
Researchers at Arctic Wolf discovered that attackers are hosting these AntiFox instances on Alibaba Cloud. From this infrastructure, they have targeted internet-facing web services and associated Amazon Web Services (AWS) environments. The attackers are not just guessing passwords; they are using valid AWS identities, such as IAM users and assumed EC2 roles, to make authenticated requests to APIs.
One specific intrusion involved the deployment of an ASPX webshell on an internet-exposed IIS server. Once the webshell was active, the attackers were able to execute operating system commands and access sensitive application settings. They specifically looked for database connection strings and copied configuration and log files to a directory that was accessible over the web.
The use of valid credentials suggests that the attackers are successfully moving laterally through systems. By obtaining initial access through a web vulnerability, they can then use the AntiFox agents to hunt for higher-value secrets. This combination of web exploitation and automated cloud discovery creates a high-speed pipeline for data theft.
Rapid credential harvesting
The speed at which these AI agents can operate is a primary concern for security teams. In one observed case, the attackers cycled through multiple identities and made hundreds of API requests in roughly 30 minutes. While many of these attempts were blocked by security measures, the attackers were still successful in some instances.
The ultimate goal of these automated bursts is often the harvesting of production secrets. In one instance, the attackers successfully obtained secrets that gave them deeper access to the environment. In a separate incident, they used a valid AWS identity to perform a series of actions on S3 objects, including reading, copying, and deleting them.
This behavior demonstrates how agents can be programmed to
Defense and detection strategies
To counter these automated threats, Arctic Wolf provides specific hunting guidance for security teams. Defenders are encouraged to correlate activity from IIS webshells with unusual access patterns within AWS environments. This helps identify when a web breach is being used as a jumping-off point for cloud exploitation.
Security professionals should also review successful API calls that occur within rapid bursts of denied requests. Because the AI agents can cycle through many identities quickly, looking for these patterns of high-frequency attempts can reveal an ongoing automated attack. Experts recommend revoking affected sessions and rotating any credentials exposed in application files.
Tool lineage and availability
AntiFox is part of a broader trend of agentic frameworks originating from China. Its predecessor, ARTEX, was linked by CrowdStrike to a cyberespionage campaign by the group Vapor Panda, which targeted over 200 organizations across 14 countries. While ARTEX went closed-source after being used against banks, AntiFox remains open-source and available on GitHub.
The platform recently reached version 2.0, which includes improvements for how the tool documents and reports its findings. Although the publishers state the tool should not be used for illegal activity or against live systems, they do not provide any technical controls to enforce these rules. This lack of restriction allows the tool to remain accessible for various types of use cases.
Sources and links
- Codegithub.com/AntiFox-AI/antifox/blob/main/README.en.md github.com
- CodeArctic Wolf says github.com
- Reportingbleepingcomputer.com: the article we sourced
First reported by bleepingcomputer.com. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.