The software industry has taken a major step forward in the fight against digital threats after Microsoft officially debuted its first specialized cyber-defense model, MAI-Cyber-1-Flash, powered by the Perception agent platform. The new technology arsenal is designed to fully automate the detection of critical vulnerabilities in program code and provide a reliable shield for corporate networks against increasingly sophisticated attacks driven by algorithms, reports TechCrunch.
The newly created model is designed specifically to analyze large-scale code bases and identify hard-to-detect vulnerabilities. It works in tight synchronization with the MDASH system, whose main function is not just to detect errors, but also to instantly apply the necessary corrections. The Perception platform builds on this process by deploying specialized teams of autonomous agents. According to Microsoft AI CEO Mustafa Suleiman, the MDASH system has demonstrated superiority in specialized Cyber Gym benchmarks, outperforming competitors such as Gemini, GPT 5.5 Cyber, GPT 5.6 Sol and Mythos 5.
Executive Vice President of Security Hayete Gallot emphasizes that cybercriminals are increasingly actively using the capabilities of artificial intelligence in their disruptions and intrusions. The Perception platform enables businesses to respond with the same speed and efficiency. The internal architecture relies on simulated operational groups - red teams for virtual staging of attacks, blue teams for identifying critical points and green teams for final patching of the code.
Lead engineer on the project Dave Weston defines the development as a fundamental leap in operational efficiency. The previously lengthy and labor-intensive manual process, carried out by dozens of specialists, is now being replaced by lightning-fast automated scanning, risk prioritization, and remediation of breaches within minutes.
Pre-test versions of MAI-Cyber-1-Flash and Perception will be available to corporate customers from November 3, setting the stage for a direct competition with similar solutions introduced this year by Anthropic and OpenAI.