Microsoft Debuts AI Security Platform Perception
AI-generated image · Bay Street Wire
The company is launching a specialized cyber model and an agentic system to automate vulnerability remediation.
Microsoft has introduced MAI-Cyber-1-Flash, its first cybersecurity-specialized model, alongside a new AI security platform called Perception, according to TechCrunch. The model is designed to identify challenging vulnerabilities within complex codebases and operates within MDASH, Microsoft's software vulnerability identification and remediation harness.
Perception utilizes agentic teams to automate security workflows. As reported by TechCrunch, the platform employs red teams for attack simulations, blue teams for bug detection and triage, and green teams to implement corrective actions. Dave Weston, lead engineer for Perception, stated the system reduces tasks that previously required hours of manual labor from specialized staff—such as remediation engineers and appsec hunters—down to minutes, providing detection, posture fixing, and code fixes.
Mustafa Suleyman, CEO of Microsoft AI, claimed that MAI-Cyber-1-Flash, when combined with GPT 5.4 in the MDASH harness, outperforms competitors including Gemini, GPT 5.6 Sol, GPT 5.5 Cyber, and Mythos 5 on the Cyber Gym benchmark. Hayete Gallot, Microsoft's vice president for security, noted that the tools are intended to help enterprises counter AI-driven attacks at scale.
Microsoft's entry arrives as other AI labs launch similar tools; TechCrunch notes that Anthropic released the Mythos platform via a program called Glasswing, and OpenAI launched a security solution in May through a program called Daybreak. Perception is scheduled to be available in preview on November 3.