Microsoft launches MAI-Cyber-1-Flash and Perception agentic security platform

Microsoft announced its first specialized AI cybersecurity model, MAI-Cyber-1-Flash, at a San Francisco event, alongside a new agent-driven security platform called Perception. According to Mustafa Suleyman, CEO of Microsoft AI, the model—when paired with GPT-5.4 inside the MDASH harness—surpasses competing models including Gemini, GPT-5.5 Cyber, GPT-5.6 Sol, and Anthropic's Mythos 5 on the Cyber Gym benchmark, the industry-standard evaluation for cybersecurity AI. The model is designed to find challenging vulnerabilities in complex codebases and integrates directly with Microsoft's MDASH remediation harness.

Perception, the new platform, deploys teams of autonomous agents categorized as red, blue, and green teams. Red teams simulate potential attacks with detailed context about threat actors and likely exploits. Blue teams detect and triage existing bugs, while green teams take corrective actions. Dave Weston, lead engineer, emphasized that Perception reduces what previously took hours of manual work from multiple security specialists down to minutes, delivering fixes, detection updates, and posture corrections. The platform enters a crowded field alongside Anthropic's Mythos (Glasswing program) and OpenAI's Daybreak. Microsoft's tools will be available in preview starting November 3, aiming to help enterprise defenders counter AI-accelerated cyberattacks with AI-powered defenses at scale.