Microsoft Unleashes AI Cybersecurity: New Model & Agentic System to Protect Digital Frontiers

Microsoft has made a significant entry into the artificial intelligence (AI) cybersecurity domain, launching its first cybersecurity-specialized model, MAI-Cyber-1-Flash, alongside a new AI cybersecurity platform named Perception. Unveiled at an event in San Francisco, these new offerings are positioned as direct competitors to established players like Anthropic, Google, and OpenAI.

MAI-Cyber-1-Flash is specifically engineered to identify challenging vulnerabilities within complex codebases. This model is integrated with MDASH, Microsoft’s dedicated harness for software vulnerability identification and remediation. The accompanying security platform, Perception, is designed to enhance and automate various security workflows by deploying specialized teams of agents. These agent teams can assist with identifying and remediating bugs and also integrate seamlessly with MDASH.

Microsoft claims that MAI-Cyber-1-Flash demonstrates superior power and cost-effectiveness compared to competitor models, a assertion supported by its performance on an established AI cybersecurity benchmark. Mustafa Suleyman, co-founder of DeepMind and current CEO of Microsoft AI, expressed excitement over the results, stating that MAI-1 Cyber Flash, when combined with GPT 5.4 within the MDASH harness, outperformed Gemini, GPT 5.5 Cyber, GPT 5.6 Sol, and Mythos 5 on Cyber Gym, which he referred to as “the primary benchmark that we all use. The golden benchmark.” Microsoft plans to ship these new tools into production immediately.

The impetus for these new solutions stems from the increasing use of AI by hackers in cyberattacks. Hayete Gallot, Microsoft’s vice president for security, highlighted Perception as a crucial tool for enterprise defenders, enabling them to “defend against AI with AI at the scale and speed that the attackers have.” The Perception platform utilizes agentic red, blue, and green teams. Red teams conduct detailed simulations of potential attacks, offering insights into threat actors and likely vulnerabilities. Blue teams focus on detecting and triaging existing bugs, while green teams are responsible for implementing corrective actions against identified vulnerabilities.

Dave Weston, the lead engineer for Perception, underscored the platform’s potential to significantly boost efficiency for corporate defenders. He noted a transformation from hours of manual work by multiple specialized security personnel to a matter of minutes for discovering issues, prioritizing them, detecting threats, fixing posture, and even generating code fixes. While AI has introduced powerful new defensive capabilities, its accessibility to cybercriminals has also spawned a complex array of new threats.

Microsoft’s new security tools are set to be available in preview on November 3. They will enter an already competitive landscape of AI cybersecurity solutions, joining offerings such as Anthropic’s Mythos, launched earlier this year through its Glasswing program, and OpenAI’s security solution, introduced in May via its Daybreak program.