Prime Highlights :
- Microsoft’s MAI-Cyber-1-Flash model is designed to detect complex vulnerabilities and powers its new MDASH system.
- Executives said the new Perception platform can reduce security workflows from hours to minutes using AI agent teams.
Key Facts :
- Microsoft is a leading technology company that develops software, cloud services and AI systems.
- The new cybersecurity tools will be available in preview starting in November.
Background :
Microsoft revealed a new AI cybersecurity platform and its first cybersecurity-specific model this week at a presentation in San Francisco.
The model, called MAI-Cyber-1-Flash, is designed to identify complex vulnerabilities in large codebases and powers MDASH, Microsoft’s system for detecting and fixing software vulnerabilities. The company also introduced a platform named Perception, built to deploy groups of AI agents that support and automate various security tasks, including bug detection and remediation. Perception can also work alongside MDASH.
Microsoft said its new model outperforms rival systems on a widely used cybersecurity benchmark, while also being more cost-effective. Mustafa Suleyman, chief executive of Microsoft AI, said the company was excited to share the results, noting that the combined model setup surpassed several competing systems on the benchmark and would be moved into production immediately.
Microsoft’s security vice president said the platform was intended to help enterprise defenders counter AI-powered cyberattacks with equally fast, AI-driven defences. Perception uses agent teams designed to simulate attacks, detect and prioritise vulnerabilities, and apply corrective fixes.
The platform’s lead engineer said the system reduced work that once took security teams hours to complete down to just minutes, covering detection, prioritisation and code fixes.
Microsoft said the new tools will be available in preview in November. The launch adds to a growing field of AI-driven cybersecurity platforms, with several major technology firms having introduced similar tools earlier this year.