Microsoft Launches Multi-Model Agent Security System
Microsoft has released a new multi-model agent security system that integrates over 100 cutting-edge and customized models to discover exploitable vulnerabilities.
The system achieved leading performance in the CyberGym benchmark tests and has helped identify and fix 16 vulnerabilities ahead of this Patch Tuesday.
Microsoft is now open for private preview registration, allowing customers to apply to test the security system.
Market Mechanism: Microsoft sells this security service to enterprise clients, driven by the demand for vulnerability fixes ahead of event-driven Patch Tuesday. Funds flow into Microsoft cloud security subscriptions and enterprise licenses, benefiting Microsoft and the Azure ecosystem while putting pressure on traditional security vendors.
Source: Public Information
ABAB AI Insight
Satya Nadella's Microsoft has previously launched agent products like GitHub Copilot and Security Copilot. This multi-model security system continues its "Copilot Everywhere" strategy from 2023, expanding from a single code assistant to a comprehensive security offensive and defensive chain, having acquired multiple security startups in 2024-2025 to enhance agent capabilities.
In terms of capital, Microsoft mobilizes Azure computing power, multi-model resources like OpenAI/Claude, and internal red team data to build a network of over 100 dedicated agent collaborations. The motivation is to enhance enterprise customer retention and Azure security revenue by discovering vulnerabilities early, while transforming security capabilities into differentiated paid services to reduce reliance on external penetration testing.
Similar cases include AI-driven vulnerability hunting by CrowdStrike and Google Mandiant, as well as Salesforce's Agentforce security expansion; Microsoft is currently transitioning from general Copilot to vertically focused high-value agents (security, operations).
Structural Judgment: This essentially represents a technology-driven restructuring of the industry chain. The multi-model agent system shifts pricing power from traditional manual red teams and single security tools to Microsoft's unified agent platform through collaborative vulnerability discovery, where a large number of dedicated agents scanning in parallel significantly reduces discovery cycles and costs, creating new entry barriers and subscription revenue pools for enterprise security.
ABAB News · Cognitive Laws
The more automated vulnerability discovery is, the more essential defense subscriptions become.
When the number of models stacks up, security capabilities become truly exponential.
Bugs fixed in-house are the most valuable moat.