Microsoft Launches AI Security Stack, Project Perception, and Cost-Effective MAI-Cyber-1-Flash Model
Microsoft introduces Project Perception, an AI-powered security platform with specialized agents, and MAI-Cyber-1-Flash, a new cybersecurity model designed for high performance at a reduced cost.
Microsoft has unveiled a significant advancement in cybersecurity with the launch of Project Perception, an AI-powered security agent stack, and MAI-Cyber-1-Flash, a novel cybersecurity model. This dual release aims to provide customers with a unified platform for continuous, AI-driven security surveillance and protection, eliminating the need for them to develop their own agents or contract with third-party AI providers.
Project Perception is designed to offer a comprehensive understanding of an organization's security environment through a team of specialized agents and models. It integrates Microsoft's multi-model security agent harness, MDASH, with the new MAI-Cyber-1-Flash model. According to Hayete Gallot, executive vice president of Microsoft Security, the initiative addresses the growing advantage attackers hold over enterprises, necessitating defense capabilities that operate at the same scale and speed.
The Perception platform operates using three distinct types of agents: 'Red' agents focused on identifying attack paths, 'Blue' agents responsible for defense, investigation, and risk assessment, and 'Green' agents tasked with remediation and hardening security postures. Users can interact with Perception by posing security-related questions, such as assessing vulnerabilities to specific attacks. The system then orchestrates the appropriate agents to complete the task, from analyzing code for vulnerabilities to simulating attacks and generating alerts.
Gallot emphasized that Project Perception provides broad visibility across identities, endpoints, applications, data, clouds, and AI systems, enabling customers to take action across these diverse environments. This capability is informed by decades of Microsoft's security research, threat intelligence, and operational experience, shaping how the system reasons, prioritizes, and responds to threats.
A key feature of Project Perception is its multi-model approach, allowing its agents to leverage the best-suited AI models available, whether from Microsoft, OpenAI, or other providers hosted within Microsoft's model garden. This flexibility ensures optimal performance for various security tasks.
Alongside the Perception platform, Microsoft introduced MAI-Cyber-1-Flash, its first proprietary cyber model. This small, code-intensive model is based on Microsoft's MAI-Thinking-1 LLM and trained on extensive security data. Integrated within the MDASH harness, it works alongside models like OpenAI's GPT-5.4 to deliver advanced capabilities.
Microsoft claims that MAI-Cyber-1-Flash, in conjunction with GPT-5.4, significantly outperformed other leading cyber-capable models, including Mythos 5 and GPT-5.5 Cyber, in the CyberGym evaluation. The MDASH harness achieved a score of 95.95%, substantially higher than its competitors. Furthermore, Microsoft AI CEO Mustafa Suleyman highlighted that the new model offers this superior performance at approximately half the cost of competing solutions, although specific pricing details were not disclosed.
This launch represents Microsoft's strategic commitment to leveraging AI for defensive cybersecurity, aiming to equip organizations with more automated, efficient, and cost-effective tools to combat the evolving threat landscape.