Microsoft showed off its first security model and intelligent agent combat system: it was tied to MDASH, leaving Gemini and GPT behind at Cyber ​​Gym. At a small event in San Francisco, Microsoft put its first dedicated network security model on the table, together with a new generation of intelligent agent security platform, and pointed the finger at An in this track.

Microsoft unveiled its first security model and agent combat system: it was tied to MDASH, leaving Gemini and GPT behind at Cyber ​​Gym. At a small event in San Francisco, Microsoft launched its first dedicated model for network security, along with a new generation of agent security platform, targeting Anthropic, Google and OpenAI in this track. This model is named MAI-Cyber-1-Flash, and Microsoft’s positioning for it is very pure—it specializes in finding difficult vulnerabilities in complex code bases. Its real stage is Microsoft's self-developed vulnerability identification and repair framework MDASH: the model is responsible for "activating" the framework, giving the matter of finding and repairing holes an engine that can run independently. The platform that debuted at the same time is called Perception. What it does is not just a single tool, but a group of agents to help enterprises automate various security processes, including discovering vulnerabilities and repairing them, and it can also be directly connected to MDASH. According to Microsoft, in this system, three types of agents, the red team, the blue team, and the green team, perform their own duties: the red team simulates potential attacks in detail, even describing the threats' portraits and the weaknesses they may exploit; the blue team specializes in detecting and sorting existing bugs; and the green team "corrects" those bugs. Mustafa Suleiman, the head of Microsoft AI, appears unabashedly excited. He said that the team tied MAI-1 Cyber ​​Flash and GPT 5.4 into the MDASH framework, and outperformed Gemini, GPT 5.5 Cyber, GPT 5.6 Sol and Mythos 5 on Cyber ​​Gym, the main benchmark recognized in the circle. "We're going into production soon," he added. Ayete Gallo, vice president of security operations, sees Perception as a weapon for defenders to use AI against AI - now that attackers have taken advantage of the speed and scale of AI, defenders have to respond with the same size. Dave Weston, the head of engineering, described this platform as a leap in enterprise defense efficiency: In the past, a group of dedicated people in the security organization-application security hunters, remediation engineers-spent hours of manual work, but now it can be done in minutes to come up with a repair plan, not only discovery and prioritization, but also detection, posture repair and even code-level patching. Widening the lens, Microsoft's move is taking place in an increasingly crowded arena. In April this year, Anthropic integrated the security platform Mythos through the Glasswing project.