Microsoft's Project Perception plans to rival Anthropic's Mythos with lower-cost AI security

Microsoft is reportedly preparing Project Perception, an AI-powered enterprise security tool designed to compete directly with Anthropic’s Mythos. The product uses a combination of AI models from Anthropic, OpenAI, and Microsoft itself to scan IT systems, identify vulnerabilities, and suggest fixes. Its key differentiator is cost: by routing individual queries to the most appropriate (and cheapest) model for each task, Microsoft aims to offer a service far cheaper than Mythos — whose API costs are estimated at 100% higher than Opus and 82% higher than GPT.

This routing approach mirrors a broader industry trend of mixing expensive frontier models with cheaper alternatives, including open-source Chinese models, to reduce operational costs. For enterprise security teams, the appeal is twofold: lower cost and integration with Microsoft’s existing ecosystem. The company is doubling down on enterprise AI under Mustafa Suleyman, who recently stated a goal to launch frontier models next year without distilling from others.

Microsoft’s push also sharpens its rivalry with OpenAI and Anthropic. CEO Satya Nadella has accused these AI companies of using customer data to improve their own models, positioning Microsoft as a more secure and governance-focused alternative. With Windows and Office already deeply embedded in enterprises, Microsoft’s pitch is that it offers better security, governance, and pricing than its partners-turned-competitors.

Geopolitics adds another layer: Anthropic faced export restrictions on its Fable 5 model (trained similarly to Mythos), limiting its availability outside the U.S. Microsoft, with its established global compliance infrastructure, may have an easier path to selling Project Perception in Europe and other regions. For businesses, the arrival of cheaper AI-driven security tools could lower the cost of closing vulnerabilities, a growing priority as bad actors also gain access to advanced AI.