Microsoft launches cybersecurity AI model and platform

Microsoft has launched MAI-Cyber-1-Flash, its first cybersecurity-specific AI model, alongside a new agentic security platform called Perception.

MAI-Cyber-1-Flash has been designed to identify vulnerabilities in complex codebases. The model powers MDASH, Microsoft’s existing harness for software vulnerability identification and remediation. According to Microsoft, the model performs competitively on Cyber Gym, an established benchmark used to evaluate AI cybersecurity models, and does so at a lower computational cost than comparable models from competitors. Moreover, the company said it is deploying the model into production immediately, combining it with GPT 5.4 within the MDASH harness.

Perception automates security workflows

Perception, the platform launched alongside the new model, is built to deploy groups of AI agents across security workflows, including bug identification and remediation. It integrates with MDASH and organises its agents into three categories: red teams, which simulate potential attacks and assess likely vulnerabilities based on threat actor behaviour; blue teams, which detect and triage existing vulnerabilities; and green teams, which carry out corrective actions once issues have been identified.

Mustafa Suleyman, the co-founder of DeepMind and current CEO of Microsoft AI, described the platform as a way for enterprise defenders to respond to AI-enabled attacks using AI-based tools at comparable scale and speed. Dave Weston, the lead engineer for Perception, said the platform consolidates tasks previously requiring multiple specialised security roles, covering detection, prioritisation, posture correction, and code remediation, into a single automated workflow.

Microsoft’s announcement comes as several major AI providers have introduced cybersecurity-focused products over recent months. Anthropic released a security platform named Mythos in April 2026 through Glasswing, a programme limited to a small number of partner organisations.

Microsoft said MAI-Cyber-1-Flash and Perception will be available in preview from 3 November 2026.

Industry implications

The launch reflects a broader shift among large AI developers toward specialised, security-focused models, as organisations face a growing volume of AI-assisted cyberattacks. While AI has expanded the defensive toolkit available to enterprises, the same technology has also broadened the range of tactics available to malicious actors, prompting security vendors to integrate agentic automation into existing detection and remediation pipelines. With this in mind, Microsoft’s entry into this space, alongside comparable moves from Anthropic and OpenAI, points to cybersecurity becoming a distinct competitive front among major AI developers, with vendors differentiating on benchmark performance, cost efficiency, and the degree of workflow automation offered to enterprise security teams.