Microsoft has unveiled two new artificial intelligence-driven tools designed to automate and streamline the identification and mitigation of security risks. The centerpiece is MAI-Cyber-1-Flash, the company's first model trained from scratch specifically for analyzing, identifying, and fixing software vulnerabilities. Built on the MAI-Thinking-1 platform, it leverages proprietary data accumulated over decades of Microsoft's patch management operations, processing over 1 trillion security signals daily and drawing insights from 1.6 million customers.
The model achieved a 96% score on the CyberGYM benchmark, outperforming competitors like Anthropic Mythos, Google Gemini, and OpenAI models—while requiring only half the cost of the previous version. Microsoft integrated MAI-Cyber-1-Flash into MDASH, a scanning framework powered by 100 AI agents working in concert.
The company also announced Project Perception, a platform that orchestrates specialized AI agents to perform red, blue, and green team functions—finding vulnerabilities, investigating risks, and applying automatic fixes. The system dynamically selects the most suitable models for each task based on cost and efficiency metrics, promising to handle up to 90% of security demands at substantially lower costs than competing market solutions. Both tools are currently available in early access for selected customers as the industry continues debating the autonomy of AI agents in critical computing environments.
