Microsoft has introduced its first AI model built specifically for cybersecurity alongside Project Perception, a platform that coordinates teams of agents to identify, assess and fix security weaknesses. The company plans to make the platform available in preview on November 3.
MAI-Cyber-1-Flash is designed to find difficult vulnerabilities in large and complex codebases. Microsoft has integrated the model into MDASH, its multi-model system for vulnerability discovery, validation and remediation.
Microsoft Reports Strong CyberGym Performance
Microsoft said an MDASH configuration combining MAI-Cyber-1-Flash with GPT-5.4 outperformed models from Anthropic, Google and OpenAI on CyberGym, a benchmark used to evaluate AI systems on cybersecurity tasks.
Microsoft AI CEO Mustafa Suleyman said the system performed better than Gemini, GPT-5.5 Cyber, GPT-5.6 Sol and Anthropic’s Mythos 5. The company also claimed that its model was more cost-effective than competing systems, although independent benchmark verification was not provided.
MDASH analyses a software repository, maps its attack surface and assigns specialised agents to investigate potentially vulnerable paths. Other agents test whether reported weaknesses can be exploited and help generate validated code fixes.
Microsoft has previously described MDASH as a system for accelerating vulnerability research across large codebases. More information about Microsoft’s existing agent-based security tools is available through its Security Copilot platform.
Project Perception Coordinates Security Teams
Project Perception expands this agent-based approach across a wider range of enterprise security work. Red-team agents simulate attacks and identify weaknesses, while blue-team agents detect, investigate and prioritise existing threats.
Green-team agents are responsible for corrective action. They can change security settings, improve an organisation’s defensive posture and generate code fixes for identified vulnerabilities.
Microsoft said the platform can combine information from identities, devices, applications, cloud services, data and AI systems. This gives its agents shared context when investigating security incidents or recommending remediation.
The platform can also integrate with MDASH, allowing vulnerabilities discovered in software to move through detection, prioritisation and remediation within the same workflow.
AI Security Market Becomes More Competitive
Microsoft said the new tools are intended to help organisations respond as cybercriminals increasingly use AI to automate attacks. Hayete Gallot, Microsoft’s vice president for security, described Project Perception as a way for defenders to operate at the same speed and scale as AI-assisted attackers.
The launch puts Microsoft in competition with AI security products from Anthropic, Google and OpenAI. Anthropic has introduced Mythos through its Glasswing partner programme, while OpenAI launched its Daybreak security initiative earlier this year.
Microsoft’s wider cybersecurity products and research are available through the company’s Security Blog and Microsoft Security website.
Featured image credits: Wikimedia Commons
For more stories like it, click the +Follow button at the top of this page to follow us.