Microsoft launches AI cybersecurity model, agentic defense platform to cut enterprise security costs

Microsoft unveiled MAI-Cyber-1-Flash, a compact in-house cybersecurity model built for vulnerability detection and remediation.

It also launched Project Perception, a multi-agent defense system using red, blue, and green agents to find, assess, and fix security issues.

Microsoft says the setup performs well on CyberGym while cutting enterprise security costs.

For especially hard cases, the orchestration layer routes tasks to OpenAI's GPT-5.4, underscoring Microsoft’s continued reliance on OpenAI for top-end reasoning.