Microsoft introduces its first AI model for cybersecurity

Microsoft has introduced its first family of AI models specifically designed for cybersecurity, along with an agentic security platform called Project Perception. The MAI-Cyber-1-Flash model was trained to perform targeted cybersecurity tasks, including analyzing software vulnerabilities and identifying attack vectors. Microsoft is taking a hybrid approach: routine tasks are handled by a compact model, while complex analyses are passed to more powerful models such as GPT-5.4. This reduces operational costs while maintaining high performance. According to the company, MAI-Cyber-1-Flash achieved 96% accuracy on the CyberGym benchmark, 12 points higher than Anthropic's Mythos. Project Perception is a platform based on multiple specialized AI agents that interact to automate incident detection, investigation, and response. It coordinates three categories of agents: Red Team for identifying vulnerabilities, Blue Team for analyzing incidents and assessing risks, and Green Team for implementing corrective measures and strengthening defenses. Microsoft is launching these solutions amid intensifying competition in the AI-powered cybersecurity market: Anthropic introduced Mythos, and OpenAI introduced Daybreak. The open preview of MAI-Cyber-1-Flash is expected in the coming weeks, while Project Perception is slated for August 3.