Microsoft launches its first cybersecurity AI model and platform

Microsoft launched its first cybersecurity-specialized AI model, MAI-Cyber-1-Flash, alongside a new agentic security platform called Perception, at a small event in San Francisco on Monday — a direct challenge to security products from Anthropic, Google, and OpenAI.

MAI-Cyber-1-Flash is built to find challenging vulnerabilities in complex codebases and powers MDASH, Microsoft’s harness dedicated to identifying and remediating software vulnerabilities. The company says the model is significantly more powerful and more cost-effective than competing models, based on its performance on an established AI cybersecurity benchmark called Cyber Gym.

Mustafa Suleyman, the co-founder of DeepMind and Microsoft AI’s current CEO, said the model — paired with GPT-5.4 inside the MDASH harness — outperforms Gemini, GPT-5.5 Cyber, GPT-5.6 Sol, and Mythos 5 on Cyber Gym, which he described as the benchmark the industry uses. He said the technology is shipping into production immediately.

Perception deploys teams of agents to assist with and automate security workflows, including identifying and remediating bugs, and it can integrate with MDASH. The platform organizes its agents into three roles: red teams that simulate potential attacks and surface likely threat actors and vulnerabilities, blue teams that detect and triage existing bugs, and green teams that take corrective action against those bugs.

Hayete Gallot, Microsoft’s vice president for security, said the goal is to let enterprise defenders “defend against AI with AI at the scale and speed that the attackers have,” pointing to hackers’ growing use of AI in their own attacks.

Dave Weston, Perception’s lead engineer, described the platform as a major efficiency gain: work that used to take hours of manual effort from specialized staff across a security organization — from appsec hunters to remediation engineers — can now be handled in minutes, with issue discovery, prioritization, detection, posture fixing, and even a code fix bundled into a single pass.

Microsoft’s new tools will be available in preview on November 3, entering an increasingly crowded field. Anthropic launched its own security platform, Mythos, earlier this year to a small group of partner organizations through a program called Glasswing, and OpenAI launched a security solution in May through a program called Daybreak.

Why it matters

With Anthropic, Google, and OpenAI all fielding competing AI security products within months of each other, cybersecurity is emerging as one of the most contested application layers in enterprise AI — and Microsoft’s efficiency claims, if borne out, could reshape how security teams are staffed.