Microsoft Launches AI Security Offensive as Corporate Breaches Mount

Microsoft launched its first dedicated cybersecurity AI model, MAI-Cyber-1-Flash, scoring 96% on the CyberGym vulnerability-finding benchmark — beating Anthropic's Mythos, Google's Gemini, and OpenAI's GPT while claiming roughly 50% cost savings. The model runs inside MDASH, Microsoft's multi-agent vulnerability harness, and anchors Project Perception, an agentic security platform that coordinates simulated red-team attacks, threat detection, and automated remediation across...