Microsoft unveils AI cybersecurity system with OpenAI and Anthropic models

Microsoft just dropped what might be the most interesting development in AI security this year. Project Perception, a new cybersecurity platform that orchestrates models from Microsoft, OpenAI, and Anthropic simultaneously, is set to enter public preview on August 3. At the center of Project Perception sits an intelligent routing system. When a cybersecurity task comes in, whether it’s vulnerability detection, threat analysis, or remediation, the platform determines which AI model is best suited to handle it. The roster includes Microsoft’s in-house MAI-Cyber-1-Flash model alongside offerings from OpenAI and Anthropic. The foundation for this approach appears to be Microsoft’s earlier MDASH system, which has been posting impressive numbers. On the CyberGym benchmark, a standardized test for AI cybersecurity performance, MDASH scored 88.45% in May 2026. That’s a meaningful gap over Anthropic’s Mythos at 83.1% and OpenAI’s GPT-5.5 at 81.8%. Microsoft is positioning Project Perception as a lower-cost alternative to single-model systems, arguing that routing tasks to the most efficient model for each job reduces overall compute spend. Microsoft revealed Project Perception on July 27, l...