Microsoft Launches Its First Cybersecurity AI Model

Microsoft has launched its first AI model built specifically for cybersecurity, alongside a new platform designed to automate security work using teams of AI agents.

The new model, called MAI-Cyber-1-Flash, is designed to find difficult security vulnerabilities in large and complex codebases.

Microsoft has integrated the model into MDASH, its system for identifying, validating, and fixing software vulnerabilities using multiple AI models and more than 100 specialized agents.

The company also introduced Project Perception, a broader AI security platform that can deploy teams of agents to monitor systems, investigate threats, and take corrective action.

Finding and Fixing Software Vulnerabilities

MAI-Cyber-1-Flash focuses on finding weaknesses in software before attackers can exploit them.

Rather than using the model alone, Microsoft runs it inside MDASH. Think of MDASH as the system that coordinates different AI models and security agents, deciding which one should handle each part of a cybersecurity task.

MAI-Cyber-1-Flash can handle up to 90% of MDASH tasks, while larger models such as GPT-5.4 can be reserved for the most difficult remaining jobs.

This approach is intended to reduce the cost of continuously scanning large amounts of software.

Microsoft Claims Better Performance at Half the Cost

Microsoft says MDASH using MAI-Cyber-1-Flash together with GPT-5.4 scored 95.95%, or roughly 96%, on CyberGym.

CyberGym is a benchmark used to test how well AI systems can examine large codebases and find real security vulnerabilities.

Microsoft says the result is around 12 percentage points higher than Anthropic’s Mythos and also ahead of systems using Gemini and GPT models.

The company also claims the new setup cuts costs by around 50% compared with its previous MDASH configuration using GPT-5.4, GPT-5.4 mini and GPT-5.3 Codex.

Microsoft AI CEO Mustafa Suleyman said the company is putting the technology into production immediately.

Project Perception Uses Teams of AI Agents

Microsoft’s second announcement, Project Perception, goes beyond finding software bugs. The platform uses AI agents to continuously monitor security risks, investigate possible attacks and respond to problems.

Microsoft divides these agents into three groups: red, blue and green teams. Red-team agents look at systems from an attacker’s perspective and search for possible ways to break in before real attackers find them.

Blue-team agents investigate suspicious activity, analyse context and decide which vulnerabilities or alerts represent genuine risks.

Green-team agents then take corrective action, such as strengthening protections or helping fix identified problems. Working together, the three groups create a continuous cycle of finding problems, evaluating them and fixing them.

Microsoft Says Fixes Can Take Minutes Instead of Hours

Dave Weston, lead engineer for Project Perception, said tasks that previously required several cybersecurity specialists and hours of manual work can now be completed much faster.

According to Microsoft, the platform can identify a vulnerability, determine how serious it is, develop detection measures, and suggest or create a code fix within minutes.

The goal is to reduce the amount of manual work required from security teams while allowing humans to remain in control of important decisions.

AI Is Also Making Cyberattacks More Powerful

Microsoft says the new tools are needed because attackers are increasingly using AI to find vulnerabilities and scale cyberattacks.

As AI makes it cheaper and faster to search large amounts of code for weaknesses, traditional security methods based on occasional scans and later patches may struggle to keep up.

Microsoft Security executive Hayete Gallot described Project Perception as a way for businesses to use AI to defend themselves at the same speed and scale as AI-powered attackers.

The company says its security systems process more than 100 trillion security signals every day across identities, devices, cloud services, networks and applications.

Microsoft Joins Growing AI Cybersecurity Race

Microsoft is entering an increasingly competitive market for specialized cybersecurity AI.

Anthropic launched Project Glasswing earlier this year, giving selected organizations access to Claude Mythos Preview for finding vulnerabilities in critical software. Anthropic later said the program had discovered more than 10,000 high- or critical-severity vulnerabilities across software tested by its partners.

OpenAI also launched Daybreak in May, combining its AI models with an agent-based system for code review, threat modeling, vulnerability detection and remediation.

Microsoft is now competing directly in the same area with MAI-Cyber-1-Flash, MDASH and Project Perception.

Project Perception will enter public preview on August 3, while MAI-Cyber-1-Flash is being deployed through MDASH for cybersecurity use.