Threat actors are increasingly leveraging AntiFox, an open-source agentic penetration testing framework, to automate credential theft and infrastructure exploitation within Amazon Web Services (AWS) environments. Security researchers at Arctic Wolf have identified malicious infrastructure hosted on Alibaba Cloud that utilizes AntiFox to conduct unauthorized activities, including vulnerability scanning, webshell deployment, and the harvesting of sensitive cloud secrets. While the developers of AntiFox maintain that the tool is intended for authorized security testing and explicitly prohibit its use in illegal or destructive operations, the software lacks technical safeguards to prevent misuse against live systems.
AntiFox originated as a fork of the ARTEX 0.1.1 framework and features a Chinese-language interface designed to facilitate the deployment of AI agents. These agents are capable of automating complex penetration testing tasks, such as mapping exposed services and probing endpoints for known vulnerabilities. The current iteration of the tool, version 2.0, includes enhanced reporting and documentation capabilities, yet it remains publicly accessible on GitHub despite its adoption by malicious actors.
In confirmed incidents, attackers have utilized Alibaba Cloud-hosted instances of AntiFox to compromise internet-facing web servers. One documented intrusion involved the deployment of an ASPX webshell on an IIS server, which allowed the unauthorized execution of operating system commands. Through this access, attackers successfully retrieved application settings, database connection strings, and various log files, often moving these sensitive documents to web-accessible directories for exfiltration.
Beyond initial server compromise, the attackers have systematically abused valid AWS identities, including IAM users and assumed EC2 roles, to interact with Amazon Web Services APIs. Arctic Wolf observed instances where attackers cycled through multiple compromised identities, generating hundreds of API requests within a 30-minute window. While many of these automated attempts were blocked by existing security controls, the attackers successfully obtained production secrets in several instances. In at least one separate case, the unauthorized use of valid AWS credentials resulted in the reading, copying, and deletion of objects stored in S3 buckets.
The emergence of AntiFox as a weaponized tool follows a broader trend of agentic AI frameworks being repurposed for cyberattacks. Previously, the ARTEX framework was linked to a significant cyberespionage campaign attributed to the group known as Vapor Panda, which targeted over 200 organizations across 14 countries. ARTEX was also implicated in attacks against South Korean financial institutions and a breach involving an internal sales application at WorkOS. Following these incidents, the developers of ARTEX transitioned the project to a closed-source model.
To mitigate the risks posed by these AI-driven attacks, security professionals are advised to implement rigorous monitoring of both web server activity and cloud API logs. Defenders should look for correlations between the deployment of web shells on internet-facing assets and unusual spikes in AWS API requests. When suspicious activity is detected, organizations should immediately revoke compromised AWS sessions and IAM keys, as well as rotate any credentials that may have been exposed through application configuration files or logs.
The use of AI agents in these campaigns highlights a shift toward highly automated, scalable exploitation tactics. By utilizing frameworks like AntiFox, attackers can rapidly probe for vulnerabilities and pivot into cloud environments with minimal manual intervention. As these tools remain available in the public domain, the security community continues to emphasize the importance of proactive threat hunting and the enforcement of the principle of least privilege for all cloud-based identities.





