Microsoft launched its first cybersecurity model and a new agentic security system in San Francisco on Monday. The move targets major rivals including Anthropic, Google, and OpenAI.
MAI-Cyber-1-Flash
The company introduced MAI-Cyber-1-Flash. It is built to find hard-to-spot vulnerabilities in complex codebases. The model runs inside MDASH, Microsoft’s harness for identifying and fixing software flaws.
Microsoft claims this model outperforms competitor versions on an established benchmark while costing less. Mustafa Suleyman, the co-founder of DeepMind and current CEO of Microsoft AI, said the team was very very excited to announce the results.
“We have MAI-1 Cyber Flash binded with GPT 5.4 inside of the MDASH harness — which beats out Gemini, GPT 5.5 Cyber, GPT 5.6 Sol, and Mythos 5 on Cyber Gym, which is the primary benchmark that we all use. The golden benchmark,” Suleyman said.
He added that the product ships into production immediately.
Perception
The new security platform is called Perception. It deploys teams of agents to assist with and automate various security workflows, including finding and fixing bugs. The system integrates with MDASH.
Hayete Gallot, Microsoft’s vice president for security, noted that hackers increasingly use AI in their cyberattacks. She described Perception as a way for enterprise defenders to defend against AI with AI at the scale and speed that the attackers have.
Perception uses agentic red teams, blue teams, and green teams. The red teams provide detailed simulations of potential attacks, including context about threat actors and likely vulnerabilities. Blue teams detect and triage existing bugs. Green teams take corrective actions against those bugs.
Dave Weston, the lead engineer for Perception, described the platform as a massive efficiency upgrade for corporate defenders. He said the process has moved from hours of manual work by multiple specialists to minutes.
“Not only do we discover the issues and prioritize them, but we have detection, posture fixing, and even a code fix,” Weston said.
The market context
Microsoft’s security tools will enter an increasingly crowded field of AI cybersecurity solutions. The company said availability in preview starts on November 3.
Earlier this year, Anthropic launched Mythos. It was released to a small group of partner organizations through a program called Glasswing. OpenAI also launched its own security solution in May through a program called Daybreak.
