HexStrike AI pentesting framework abused to exploit Citrix vulnerabilities

HexStrike AI, an open-source framework for AI-driven penetration testing, was used by cybercriminals to exploit recent Citrix vulnerabilities, Check Point reported Tuesday.

HexStrike AI v6.0, which was released about two weeks ago, can enable AI agents such as Anthropic’s Claude, OpenAI’s GPT or Microsoft’s Copilot to autonomously leverage more than 150 cybersecurity tools via its model context protocol (MCP) platform.

This setup includes a FastMCP server which serves as the “orchestration brain” for agents to communicate and call tools, an abstraction layer that converts natural language prompts into technical workflows, and retry logic and recovery handling to reliably sustain automated processes, Check Point described in its blog post.  

Through this platform, AI agents can seamlessly leverage tools for network scanning, password cracking, reverse engineering and other cybersecurity and offensive security processes in order to target specific software and vulnerabilities without manual intervention.

The framework also features more than a dozen specialized AI agents, including agents focused on exploit development, vulnerability intelligence and bug hunting workflows.

Check Point noted that while HexStrike AI can be used by cybersecurity researchers for AI-driven red teaming, penetration testing and other offensive security operations, cybercriminals have already begun leveraging the framework in the hopes of speeding up exploitation of vulnerabilities.

On cybercriminal forums, threat actors have discussed using the framework to identify and exploit instances of Citrix NetScaler ADC that are vulnerable to three vulnerabilities disclosed last week, including a critical zero-day tracked as CVE-2025-7775.

CVE-2025-7775, which had already been exploited in the wild when it was disclosed on Aug. 26, 2025, is a memory overflow flaw that could lead to unauthenticated remote code execution (RCE).

With a CVSS score of 9.2, CVE-2025-7775 is noted to have a high attack complexity, but according to Check Point, an AI-driven framework like HexStrike could reduce the time for developing an exploit from “days or weeks” to “under 10 minutes.”

“Exploiting these vulnerabilities is non-trivial. Attackers must understand memory operations, authentication bypasses, and the peculiarities of NetScaler’s architecture. Such work has historically required highly skilled operators and weeks of development. With Hexstrike-AI, that barrier seems to have collapsed,” Check Point Cyber Security and AI Expert Amit Weigman wrote.

HexStrike AI developer Muhammad Osama did not respond to requests for comment from SC Media, but HexStrike AI responded to Check Point’s blog in a LinkedIn post, stating: “The framework was built to empower defenders, red teams, and researchers, not to fuel malicious activity. Its purpose is to give security professionals an AI-powered orchestration layer to test, detect, and harden environments against the very kinds of attacks described in the article.”

The post also noted that these recent developments demonstrate that defenders must adapt to the use of AI by cybercriminals using “the same level of automation and intelligence that attackers are racing toward,” and encouraged organizations to speed up CVE patching through automation, invest in AI-driven defenses and monitor the dark web for threat intelligence.

Check Point similarly concluded its blog by recommending fast, automated vulnerability patching, threat intelligence monitoring, AI-driven defenses such as telemetry analysis, anomaly detection and automated incident response, and the use of adaptive detection technologies that go beyond static rules and signatures.