When cybersecurity researchers discover that an attack was simply quick rather than sophisticated, a certain kind of fear descends upon them. That is precisely how the CyberStrikeAI campaign, which was the subject of three different investigations in early 2026, makes one feel. More than 600 FortiGate firewalls were compromised. 55 nations. An official assessment of a threat actor from Amazon’s own team revealed that they had “low-to-medium baseline technical capability.” No days with zero. No top-tier craft. Just a human operator watching as an AI platform handles the heavy lifting.
The core of this is CyberStrikeAI, an open-source, Go-based offensive security platform created by Ed1s0nZ, a GitHub user with documented connections to China’s National Vulnerability Database and Knownsec 404. The platform uses a single AI orchestration engine to integrate more than 100 security tools, including Nmap, Masscan, SQLmap, Metasploit, Mimikatz, and many more.
Enter a natural-language prompt into the dashboard, and the system takes care of the rest—credential harvesting, scanning, and chaining findings from one tool into another—without requiring the operator to comprehend the inner workings of the system. An attacker could launch a worldwide intrusion campaign from a Chinese enterprise messaging app on their phone because it even integrates with DingTalk and Lark chatbots.
Beginning in February 2026, Amazon Threat Intelligence discovered a campaign that targeted management ports on FortiGate appliances in over 55 countries for five weeks, from January 11 to February 18. 443, 8443, 10443, and 4443 are TCP ports. admin interfaces that are visible. authentication using a single factor. Exotic methods are not necessary. Lateral movement planning, configuration extraction, and brute-force credential attempts were all managed by the AI. According to Amazon, the human operator had trouble compiling exploits and “creative problem-solving during live operations.” CyberStrikeAI filled that gap, so it didn’t matter.
The involvement of Team Cymru added a level of technical accuracy that is difficult to overcome. Team Cymru’s Scout platform discovered a CyberStrikeAI service banner operating on port 8080 after Amazon shared one of the campaign’s command-and-control IPs, 212.11.64[.]250. Direct communication between that server and the FortiGate targets that Amazon had previously identified was verified by NetFlow analysis.
Team Cymru monitored 21 distinct attacker-controlled CyberStrikeAI servers, mostly located in China, Singapore, and Hong Kong, between January 20 and February 26. After delving even further into the exposed infrastructure, Cyber and Ramen, an independent research team, discovered over 1,400 files spread across 139 subdirectories of operational artifacts. Additionally, CHECKER2 logs revealed that over 2,500 FortiGate appliances were in line for automated scanning across more than 100 countries. These weren’t verified breaches; they were targets in the pipeline. Even so, it’s worth considering the implications of that figure for ambition.
It’s possible that the architecture that made the breach possible rather than the breach itself is what makes this story unsettling. CyberStrikeAI’s agent can orchestrate external services beyond its built-in toolkit by utilizing Anthropic’s Model Context Protocol, a standardized interface between AI models and external tools that was introduced in late 2024. With just one configuration file modification, the platform can switch between DeepSeek, GPT-4o, and Claude as AI backends. The end product is a system that can be downloaded, configured, and deployed globally against enterprise infrastructure by any actor with a modicum of motivation at a scale that previously required coordinated human teams.
Here, a more general pattern is emerging. AI-enabled attacks increased by 89% year over year, and autonomous AI agents are now responsible for about one in eight AI-related breach events, according to aggregated threat intelligence data from early 2026. The CyberStrikeAI campaign fits into that pattern rather than being an anomaly; rather, it is more akin to a proof of concept that performed better than anyone was willing to acknowledge. The impacted businesses weren’t using obscure, outdated systems. FortiGate firewalls are a common type of enterprise infrastructure. The exploited vulnerabilities weren’t brand-new. Weak credentials and exposed management interfaces have long been on every security checklist.
Above all, CyberStrikeAI highlights the disconnect between knowing what needs to be fixed and actually fixing it. Businesses are aware of the dangers associated with single-factor authentication on admin portals that are accessible online. In any case, many haven’t fixed it. AI doesn’t solve that failure; rather, it makes it far more costly to ignore. When AI is performing the scanning, the exploit window—which was previously measured in days rather than months for the majority of common vulnerability classes—may now be measured in hours. Defenders continue to use patch cycles designed for a different time period.
When examining this campaign through the prism of three distinct investigations, it’s difficult to ignore the fact that what the tool reveals about the state of enterprise security hygiene at scale is more concerning than the tool itself. CyberStrikeAI did not innovate. It passed through open doors.
