Microsoft Readies Project Perception: Multi-Model AI Cybersecurity to Rival Anthropic Mythos

Microsoft is preparing to ship an AI cybersecurity product codenamed "Project Perception" — and its defining feature is that it does not bet on a single model.

Three Key Takeaways

Multi-model routing architecture. Project Perception plugs into models from Microsoft, OpenAI, and Anthropic simultaneously, using a smart routing layer to match each security task to the best-fit model. Vulnerability scanning might run on one model, code auditing on another, threat analysis on a third — whichever is cheapest and most accurate wins.

Directly targeting Anthropic Mythos. Microsoft is explicitly positioning Project Perception against Anthropic's Mythos security product. Reports indicate Mythos API costs run roughly 100% above Opus and 82% above GPT; Microsoft's play is to undercut on price while offering deeper enterprise ecosystem integration.

Launching this month. The project is led by Hayete Gallot, who became Microsoft's security chief in February 2026, and is expected to ship in July. It will integrate tightly with Microsoft's existing security stack — Defender, Sentinel, Entra — creating a detect-to-fix closed loop.

WangDou's Take

This is textbook Microsoft "platform tax" strategy: you do not pick the model, you do not build the pipeline, you do not handle compliance — we do it all, and cheaper than Anthropic. Sounds great on a slide deck. But think about what it actually means: your security data flows through three different AI providers' models. For companies that will not even let source code touch GitHub Copilot, the trust bar for this setup is actually higher than a single-vendor solution. Microsoft is betting that most CISOs will not think that hard — they just want an "AI security solution" they can put in a procurement report. And "Microsoft" is, as always, the three most powerful letters in enterprise purchasing.

Source: TechRepublic, Neowin