Microsoft has launched its first artificial intelligence model built specifically for cybersecurity, alongside a new agentic security platform designed to automate vulnerability detection and strengthen enterprise cyber defenses. The announcement marks Microsoft's latest push into AI-powered security as cyber threats become increasingly sophisticated and AI-assisted attacks continue to rise.
The new model, named MAI-Cyber-1-Flash, has been developed to identify complex software vulnerabilities and assist security teams in discovering and remediating flaws more efficiently. Rather than functioning as a general-purpose large language model, it has been trained specifically for cybersecurity tasks and is integrated into Microsoft's existing vulnerability discovery framework known as MDASH, a multi-agent system that orchestrates AI agents to identify, validate and prioritize security issues.
Alongside the model, Microsoft introduced Project Perception, a new agentic cybersecurity platform that combines AI models, enterprise security signals and specialized software agents to continuously identify risks and recommend or execute defensive actions. The company said the platform is designed to help organizations shift from reactive security operations to continuous AI-assisted cyber defense. Project Perception will enter public preview beginning August 3.
According to Microsoft, MAI-Cyber-1-Flash has been optimized to handle around 90% of routine vulnerability analysis tasks, while more computationally intensive cases can be escalated to larger foundation models when necessary. The company said this hybrid architecture reduces infrastructure costs while maintaining high performance. Microsoft claims the new configuration delivers approximately 50% lower operating costs compared with previous multi-model cybersecurity systems.
The company also reported strong benchmark performance for its new model. When deployed within the MDASH framework, MAI-Cyber-1-Flash achieved a score of nearly 96% on the CyberGym vulnerability benchmark, outperforming several competing AI cybersecurity models in Microsoft's internal evaluations. The company said the system combines more than 100 specialized AI agents that collaborate to identify, validate and prioritize vulnerabilities across software environments. These benchmark results have been published by Microsoft and have not been independently verified.
Microsoft executives said the rapid evolution of AI has fundamentally changed the cybersecurity landscape. Attackers are increasingly using AI to accelerate reconnaissance, vulnerability discovery and exploit development, requiring defenders to adopt equally advanced AI-driven tools. The company said Project Perception has been designed to continuously perceive risks, reason across large volumes of security data and take action at machine speed while keeping security professionals in control of final decisions.
The launch comes as competition intensifies among technology companies developing AI-powered cybersecurity platforms. Google, Cisco and several specialist security firms have introduced AI agents for vulnerability management and threat detection over the past year. Rather than relying on a single foundation model, Microsoft is positioning its platform around an orchestration approach that combines specialized AI models with autonomous agents for different security workflows.
Industry analysts view specialized AI models as the next stage of enterprise AI adoption, particularly in cybersecurity where domain-specific knowledge and speed are critical. While AI agents are expected to improve operational efficiency, many organizations continue to balance automation with human oversight, particularly for high-risk security decisions. Microsoft's latest launch reflects this broader industry trend of combining agentic AI with expert supervision rather than fully autonomous cybersecurity operations.
As enterprises continue expanding AI deployments across their technology infrastructure, Microsoft's introduction of a dedicated cybersecurity model underscores the growing focus on building AI systems designed not only to enhance productivity but also to defend against increasingly sophisticated digital threats.