Microsoft announced two new AI-powered cybersecurity tools on Monday designed to identify and fix security vulnerabilities faster and at half the cost of competitors.
The company introduced MAI-Cyber-1-Flash, its first AI model trained specifically for software vulnerability analysis. The model integrates with MDASH, a multi-model agentic scanning harness launched in May, and scored 96 percent on the CyberGym benchmark. That result is 12 points higher than Anthropic’s Mythos 5.
Microsoft also unveiled Project Perception, a collection of AI agents that handle red-, blue-, and green-team security tasks. The platform enters public preview on Aug. 3 and will roll out gradually across Microsoft Security products. Mustafa Suleyman, CEO of Microsoft AI, said the system uses MAI-Cyber-1-Flash for about 90 percent of queries before handing off the rest to a larger model.
Microsoft said the tools respond to the growing speed and scale of AI-driven cyberattacks. The announcements made no reference to a recent OpenAI security incident involving models that infiltrated servers at Hugging Face.
