CyberStrikeAI: The Open-Source AI Pentest Platform Now Fueling Real Attacks

Source: ИИ-инструмент для пентеста CyberStrikeAI уже используется хакерами — Xakep (xakep.ru), original language: Russian. Translated & adapted by Sourav Banerjee.

Original URL: https://xakep.ru/2026/03/05/cyberstrikeai/

Researchers at Team Cymru have found that the open-source AI platform CyberStrikeAI is being used by attackers to conduct automated attacks. The tool featured in a recent malicious campaign that compromised hundreds of Fortinet FortiGate firewalls.

Background: the FortiGate campaign

Last month we covered how a Russian-speaking attacker compromised more than 600 FortiGate firewalls across 55 countries in five weeks using generative AI tools. Now, Team Cymru researcher Will Thomas (aka BushidoToken) found that on one of the servers involved in that campaign (212.11.64[.]250), the CyberStrikeAI tool was running.

NetFlow analysis showed a CyberStrikeAI service banner on port 8080 at that address, with active network exchange between that IP and the target FortiGate devices. The platform was last observed in this infrastructure on January 30, 2026.

What CyberStrikeAI is

Per its GitHub description, CyberStrikeAI positions itself as an "AI platform for security testing," written in Go. It bundles:

• 100+ security tools
• An orchestration engine
• Pre-installed roles
• A skills system

"Thanks to native support for the MCP protocol and AI agents, the platform provides end-to-end automation — from natural-language commands to vulnerability discovery, attack-chain analysis, knowledge extraction, and result visualization," the project description reads.

The AI decision engine is compatible with models such as GPT, Claude, and DeepSeek. The open-source platform also offers a password-protected web interface with action logging, SQLite storage, and a dashboard for vulnerability management, task orchestration, and attack-chain visualization.

Its arsenal enables full-scale attacks:

• Network scanning: nmap, masscan
• Web application testing: sqlmap, nikto, gobuster
• Exploitation: Metasploit, pwntools
• Password cracking: hashcat, john
• Post-exploitation: mimikatz, bloodhound, impacket

Combining these tools with AI agents and the orchestrator lets even inexperienced users automate attacks on targets. Team Cymru warns such solutions can noticeably accelerate automated attacks on edge devices, including firewalls and VPN gateways.

Observed infrastructure

Between January 20 and February 26, 2026, researchers identified 21 unique IP addresses running CyberStrikeAI. Servers were located mainly in China, Singapore, and Hong Kong, with additional infrastructure in the US, Japan, and Europe.

The developer and the attribution angle

Researchers also examined the platform's developer profile (Ed1s0nZ), who created other AI tools:

• PrivHunterAI — for privilege-escalation vulnerability discovery
• InfiltrateX — a privilege-escalation scanner

Notably, the developer's GitHub activity suggests potential ties to organizations previously linked to Chinese government cyber units. For example, in December 2025 Ed1s0nZ shared CyberStrikeAI with the Starlink project of Knownsec 404 (a Chinese infosec company). In January 2026 the developer mentioned receiving an award from CNNVD (China National Vulnerability Database) — believed to be run by the Chinese intelligence community. That information was later removed from the profile.

Takeaway

This is a textbook case of the dual-use problem on the offensive AI side: a purpose-built, open-source platform that accelerates authorized security testing is equally effective at lowering the skill bar for attackers. The FortiGate campaign shows boundary devices (firewalls, VPN gateways) are prime targets for this kind of automation — they sit at the network edge and their compromise grants broad access. Defenders should treat any internet-visible CyberStrikeAI instance as malicious infrastructure, and harden edge devices against AI-orchestrated scanning and exploitation that can move far faster than human-led attacks.