AntiFox AI pentesting tool linked to AWS credential theft campaigns
Arctic Wolf Labs reports that the open-source AntiFox AI platform is being used to steal AWS secrets and pivot through compromised IIS servers.
Security researchers at Arctic Wolf Labs have identified a campaign where threat actors are using AntiFox, an open-source AI-driven penetration testing platform, to steal Amazon Web Services (AWS) credentials. The attacks, observed in recent intrusions, involve chaining web exploitation with automated cloud enumeration to access production secrets and move laterally within compromised environments.
What happened
The malicious activity centers on the misuse of AntiFox, a GitHub-hosted project described as an autonomous penetration testing agent. As of the report, the repository has been starred 474 times and forked 251 times. While the tool is designed for authorized security engagements, attackers have deployed it from Alibaba Cloud systems to target internet-facing web services and their associated AWS infrastructure. Arctic Wolf Labs noted that these systems were directly associated with confirmed malicious actions.
In one specific attack vector, adversaries gained initial access through a compromised Internet Information Services (IIS) application. Although the precise vulnerability remains unknown, the attackers leveraged this access to deploy an ASPX web shell via the IIS worker process, w3wp.exe. This allowed them to execute operating system commands, locate configuration files such as web.config, and extract database connection strings. The attackers then used these stolen credentials to attempt connections to internal production databases and copied sensitive logs into web-accessible directories.
In other instances, the threat actors began with valid AWS credentials obtained through undetermined means. Using these identities, they enumerated dozens of AWS services, including IAM, EC2, S3, Lambda, Secrets Manager, and Amazon Bedrock. While many API requests resulted in AccessDenied errors, the attackers successfully retrieved values from production secrets in Secrets Manager and performed object operations in S3 buckets. This demonstrates how automated tools can rapidly validate credentials and identify high-value targets across a complex cloud environment.
How it works
AntiFox operates using a multi-agent system comprising a planner, worker, and main agent. These components interact with a dual-graph core that maintains a global asset graph and a per-task exploration graph. The workflow follows a structured path: reconnaissance, attack-surface discovery, exploitation, finding capture, and report delivery. The project maintainer emphasizes that unlike traditional tools that stop at discovery, AntiFox focuses on delivering actionable evidence and reproducible findings.
The weaponization of this tool highlights a shift in how agentic frameworks are applied. Instead of manually directing each step of an intrusion, operators set a high-level goal and allow the AI agents to execute the necessary steps autonomously. The agent can connect to large language models to assemble pipelines at runtime, downloading scanners, searching GitHub for secrets, and passing keys between tools without human intervention. This automation allows attackers to scale their efforts and maintain persistence with minimal manual oversight.
Key details
- AntiFox is an open-source AI penetration testing platform hosted on GitHub with 474 stars and 251 forks.
- Attackers used Alibaba Cloud systems hosting AntiFox interfaces to target internet-facing web services and AWS environments.
- Initial access in some cases involved compromising IIS applications and deploying ASPX web shells via w3wp.exe.
- Stolen credentials from web.config files were used to access internal production databases and copy logs to web-accessible directories.
- Valid AWS identities were abused to enumerate services like IAM, S3, Lambda, and Secrets Manager, leading to successful retrieval of production secrets.
- The tool’s multi-agent system uses a dual-graph core to manage reconnaissance, exploitation, and reporting autonomously.
Why it matters
For software engineers and security teams, this incident underscores the risk posed by accessible AI-powered offensive tools. Agentic frameworks like AntiFox lower the barrier to entry for sophisticated attacks, allowing threat actors with limited technical skills to execute complex intrusion chains. The speed and consistency of these automated attacks mean that vulnerabilities which might be overlooked or deprioritized during manual testing can be exploited rapidly and at scale.
The ability of these tools to pivot from web application compromises to cloud infrastructure enumeration creates a significant blast radius. A single misconfigured web server or leaked credential can lead to full compromise of production secrets and data stores. Security teams must assume that attackers are using automated agents to validate credentials and explore cloud permissions continuously, making static defense measures insufficient.
What you can do
- Audit IIS applications for unauthorized web shells and monitor w3wp.exe for unusual command execution patterns.
- Restrict IAM permissions to follow the principle of least privilege, ensuring that compromised credentials cannot enumerate all AWS services.
- Implement strict monitoring for Secrets Manager and S3 API calls, alerting on unusual enumeration or GetSecretValue requests.
- Review and rotate database connection strings and application settings stored in web.config files regularly.
- Monitor for exposed consoles of open-source pentesting tools like AntiFox and ARTEX in your external attack surface.
- Enforce multi-factor authentication and robust identity governance to prevent the misuse of valid AWS credentials.



