Microsoft Unveils MAI-Cyber-1-Flash AI Model for Cybersecurity, Claims 96% Benchmark Score

Microsoft has introduced MAI-Cyber-1-Flash, its first in-house AI model designed specifically for cybersecurity, alongside a new autonomous security platform called Project Perception. The announcements were made during a security event in San Francisco.

According to Microsoft, MAI-Cyber-1-Flash, when paired with OpenAI’s GPT-5.4 inside the company’s MDASH vulnerability management platform, achieves 96% accuracy on the CyberGym benchmark—reportedly outperforming competing cybersecurity AI models while operating at 50% lower cost than Microsoft’s previous MDASH configuration.

Built for Automated Cyber Defense

The company says MAI-Cyber-1-Flash is designed to handle around 90% of routine cybersecurity tasks, allowing GPT-5.4 to focus only on more complex investigations and decision-making.

The model is based on Microsoft’s MAI-Thinking-1 architecture and is trained using insights from more than 100 trillion daily security signals collected across:

• Identity security
• Endpoint protection
• Cloud infrastructure
• Enterprise networks

Microsoft AI CEO Mustafa Suleyman said the new system delivers “world-leading performance at 50% of the cost.”

Introducing Project Perception

Microsoft also announced Project Perception, an agentic cybersecurity platform that coordinates multiple AI agents to automate security operations.

The platform consists of three specialized AI teams:

• Red Team Agents – Simulate attacker behavior and identify potential attack paths.
• Blue Team Agents – Detect, investigate, and prioritize active cyber threats.
• Green Team Agents – Automatically remediate vulnerabilities and implement security fixes.

According to Microsoft engineer Dave Weston, Project Perception can reduce security workflows that previously required hours of manual effort by multiple specialists into just a few minutes.

The platform is scheduled to enter public preview on August 3, 2026.

Expanding Microsoft’s AI Security Portfolio

The launch follows Microsoft’s broader strategy to build more AI models internally. In June, the company introduced a family of seven proprietary AI models, including MAI-Thinking-1, reducing its reliance on external AI providers while expanding enterprise AI offerings.

With cybersecurity remaining a key business segment, Microsoft continues investing heavily in AI-powered defense technologies. The company previously reported annual cybersecurity revenue exceeding $20 billion, and the new tools position it to compete with specialized security AI offerings from Anthropic and OpenAI.