Microsoft Unveils Its First Cybersecurity-Focused AI Model and Agentic Security System

In an effort to keep up with AI rivals, Microsoft released its first cybersecurity model, MAI-Cyber-1-Flash, on Monday alongside a new security system called Project Perception. Microsoft says the former is designed to find software vulnerabilities and help with fixing them within security processes, not to be a general-purpose AI.

MAI-Cyber-1-Flash does most of the work in MDASH, the company’s multi-model security scan. Microsoft routes the hardest cases to GPT-5.4, which it says helps cut costs by about 50% compared with its previous best configuration. The new model reportedly achieved a score of 95.95% on the CyberGym benchmark, which tests its ability to reproduce software vulnerabilities. That’s 12 points ahead of Anthropic’s Mythos, OpenAI’s GPT-5.6 Sol, and Google’s Gemini 3.5 Flash Cyber.

Project Perception adds an extra layer to Microsoft’s security framework. It uses various AI agents—color-coded red, blue, and green—to spot and resolve security gaps. Red team agents identify paths to compromise, blue team agents investigate and assess risk, and green team agents take corrective actions. Microsoft emphasizes that despite these agents’ flexibility, human approval is required for major software changes.

Microsoft revealed MDASH back in May and is expanding it this upcoming Monday, August 3. Project Perception will be available for public preview then, starting with early access via Microsoft Defender and later expanding to other Microsoft Security products.