Microsoft has launched its first artificial intelligence model designed specifically for cybersecurity, alongside a new agent-based security platform aimed at helping organizations detect vulnerabilities and defend against increasingly sophisticated AI-powered cyberattacks.
The announcements were made during a company event in San Francisco on Monday.
The new model, called MAI-Cyber-1-Flash, has been developed to identify complex software vulnerabilities and works alongside Microsoft's MDASH framework, which focuses on software vulnerability detection and remediation.
Microsoft said the model is designed to help security teams find difficult-to-detect flaws in large codebases more efficiently.
Microsoft challenges AI rivals
Microsoft claimed its new cybersecurity model outperformed competing AI systems from Anthropic, Google, and OpenAI on the widely used Cyber Gym benchmark.
Mustafa Suleyman, CEO of Microsoft AI and co-founder of DeepMind, said the company had combined MAI-Cyber-1-Flash with GPT-5.4 inside the MDASH framework to achieve the results.
"We're very excited to announce our results," Suleyman said.
According to Microsoft, the model outperformed competing systems, including Gemini, GPT 5.5 Cyber, GPT 5.6 Sol and Anthropic's Mythos 5.
The company said the technology would be deployed into production immediately.
New AI platform automates cyber defence
Alongside the model, Microsoft introduced Perception, a new AI-powered cybersecurity platform that deploys teams of autonomous AI agents to perform different security tasks.
The platform is designed to automate security workflows such as vulnerability detection, prioritization and remediation while integrating directly with MDASH.
Microsoft said Perception uses three categories of AI agents:
- Red teams simulate cyberattacks, identify potential vulnerabilities and assess likely threat actors.
- Blue teams detect, investigate and prioritize existing security issues.
- Green teams automatically implement corrective actions, including generating code fixes.
AI defending against AI
Hayete Gallot, Microsoft's vice president for security, said cybercriminals are increasingly using artificial intelligence to launch sophisticated attacks, making AI-powered defence essential.
"We need to defend against AI with AI at the same scale and speed attackers are operating," she said.
Dave Weston, Microsoft's lead engineer for Perception, said the platform significantly reduces the time required to identify and fix vulnerabilities.
"What previously took hours of manual work from multiple security specialists can now be completed in minutes," Weston said.
The system not only identifies security flaws but also prioritizes them, recommends fixes, updates security posture and can even generate code patches automatically, he added.
Competition intensifies
Microsoft's latest products enter an increasingly competitive AI cybersecurity market.
Earlier this year, Anthropic introduced its Mythos cybersecurity platform through its Glasswing programme, while OpenAI has also expanded its cybersecurity capabilities.
Microsoft said both MAI-Cyber-1-Flash and the Perception platform will be available in preview beginning November 3.
