[DigitalToday reporter Chi-gyu Hwang] Microsoft has unveiled its first cybersecurity-focused AI model and an agent-based security platform.

TechCrunch reported on July 27 that Microsoft introduced the security model "MAI-Cyber-1-Flash" and the platform "Perception" at an event in San Francisco, California.

"MAI-Cyber-1-Flash" is designed to find difficult vulnerabilities in complex codebases. It will be embedded in Microsoft’s "MDASH" project for identifying and fixing software vulnerabilities. Microsoft said that based on existing AI cybersecurity benchmarks, the model outperforms rival models on performance and cost efficiency.

Mustafa Suleyman (무스타파 술레이만), chief executive of Microsoft AI, said that within MDASH, "MAI-Cyber-1-Flash" combined with GPT 5.4 outperformed Gemini, GPT 5.5 Cyber, GPT 5.6 Sol, Mythos 5 and "Mitos 5" on the "Cyber Gym" benchmark.

"Perception" is a platform that supports and automates security work by deploying multiple agents. It handles various security workflows such as identifying and fixing bugs, and also links with MDASH.

The Perception platform consists of red team, blue team and green team agents. The red team simulates potential attacks, while the blue team detects and classifies existing bugs. The green team carries out remediation measures for those bugs. Hayete Gallo (하예테 갈로), Microsoft’s vice president for security, said it is a way to help corporate defenders respond with AI at the same scale and speed as attackers as hackers increasingly use AI.

Dave Weston (데이브 웨스턴), who leads Perception, stressed that the platform can cut manual work that multiple specialists used to spend hours on down to minutes. He said it can not only find issues and set priorities, but also handle detection, security posture remediation and code fixes.

Microsoft’s new security tools are due to be offered in preview on Nov. 3. Earlier, Anthropic unveiled the security platform Mythos to some partners through its "Glasswing" programme early this year, and OpenAI has also been providing a security-focused AI model since May under its "Day Break" programme.