Microsoft is preparing to launch Project Perception, a new AI-powered cybersecurity platform that aims to make continuous vulnerability discovery more accessible and affordable for enterprises. Expected to debut before the end of July, the platform uses multiple AI models, from OpenAI to Anthropic, to scan software for exploitable weaknesses while reducing the cost traditionally associated with frontier AI security tools.
The vulnerability discovery offering comes after the company recently used AI to help identify vulnerabilities addressed in its July Patch Tuesday release, which fixed a record 570 security flaws.
This is Microsoft's clearest attempt to use its AI capabilities to enter the burgeoning AI cybersecurity market, positioning it as a leading rival to the much-talked-about Claude Mythos and Fable 5. However, the details of the launch suggest a different entry point for enterprises looking at AI security.
Microsoft Is Betting on Smart Orchestration
Project Perception has been designed to lower the cost of continuous code scanning by reserving expensive frontier AI models for the most complex reasoning tasks while assigning routine analysis to smaller, less costly models. Tasks such as inventory checks, log analysis, and initial triage can be completed using distilled models before escalating only the most difficult security questions to more capable systems.
That routing layer is underpinned by Microsoft's work on MDASH, the Multi-model Agentic Scanning Harness. The system orchestrates more than 100 specialized AI agents that prepare codebases for analysis, identify potential vulnerabilities, validate findings, remove duplicate results, and ultimately attempt to prove whether a vulnerability is genuinely exploitable.
Microsoft has steadily demonstrated MDASH's progress throughout 2026. The platform achieved a 96.55% score on the CyberGym benchmark after significant improvements over previous versions, while retrospective testing across audited Windows components reportedly identified nearly all previously confirmed security flaws. The technology has also been integrated into Microsoft's internal engineering workflows across Windows, Azure, and identity systems, giving the company a foundation on which to commercialize the platform.
The launch also reflects Microsoft's broader AI security strategy. Rather than building a platform around a single flagship model, Microsoft is focusing on creating an orchestration layer capable of combining the strengths of multiple AI systems into a single security workflow. By doing so, Microsoft aims to deliver enterprise-grade vulnerability discovery at a significantly lower cost than Anthropic charges for direct Mythos access.
How Does It Compare With Anthropic's Mythos?
Project Perception enters a market where Anthropic's Claude Mythos Preview has become the standard against which every AI vulnerability discovery platform is likely to be measured. Mythos demonstrated that frontier AI can autonomously identify exploitable software vulnerabilities at a scale beyond traditional security testing, but its availability has remained tightly controlled, either through Anthropic's invitation-only Project Glasswing program or via a more restricted version of the model with Fable 5.
Microsoft's strategy differs in a fundamental way. Rather than attempting to build a direct replacement for Mythos, Project Perception combines multiple AI models, including Anthropic's own, and intelligently routes workloads between them. The goal is to deliver frontier-level security analysis while avoiding the cost of running the most advanced models across every stage of the scanning process. This is especially important as vulnerability discovery becomes more frequent to keep pace with attackers using AI to identify vulnerabilities at greater speed and scale.
