Microsoft has launched Project Perception, an agentic security system for organisations, and introduced Microsoft AI-Cyber-1-Flash, its first AI model focused on cyber security. Project Perception is designed to continuously identify, assess and reduce security risk across an organisation's environment, organised around three groups of agents: Red agents identify possible attack paths and software weaknesses, Blue agents investigate findings to determine material risk, and Green agents take corrective action, forming a closed-loop process. Alongside the system, Microsoft unveiled Microsoft AI-Cyber-1-Flash, a cyber security model built for software vulnerability analysis, introduced first in a software vulnerability management workflow inside MDASH. According to Microsoft, the combination of MDASH and Microsoft AI-Cyber-1-Flash scored 96% on the CyberGym benchmark, 12 points above Mythos, while cutting costs by almost 50% compared with the current MDASH configuration on the market. Project Perception is built to select models based on quality, reliability, latency and cost. The launch also reflects Microsoft's effort to define a new cyber stack for AI-era defence, intended to create a continuously learning security system.