Source: Microsoft plans to release an AI security tool this month that uses models from Anthropic, OpenAI, and itself, as a cost-effective alternative to Mythos
Microsoft is preparing to release a new AI security product, internally codenamed Project Perception, to capture a piece of companies' rising cyber defense spending.
Context & Ripple Effects
Related coverage shows Microsoft simultaneously deepening its use of Anthropic models and working to replace some third-party model usage with its own MAI models to lower costs. Anthropic’s Mythos has also been tested by the NSA for finding software vulnerabilities, giving the new product a clear comparison point in a security-sensitive use case.
Project Perception therefore extends Microsoft’s model-sourcing strategy into cyber defense: it can package external and internal models around an enterprise security workflow while competing on the cost of the resulting task.
First-order effects
- Microsoft gains a planned security offering aimed at corporate cyber-defense budgets and positioned directly against Mythos on cost.
- Anthropic and OpenAI become both model suppliers to Microsoft’s product and, in Anthropic’s case, adjacent competitors where the product overlaps with Mythos.
Second-order effects
- Microsoft can use a mix of Anthropic, OpenAI, and MAI models to manage performance and inference costs, reducing the need for any one supplier to power the entire workflow.
- Mythos and other AI-security offerings face pressure to demonstrate differentiated vulnerability-finding performance or pricing rather than relying on frontier-model access alone.
Third-order effects
- If enterprises adopt multi-model security products, value may concentrate in the vendor that owns the security workflow, customer relationship, and cost controls—not necessarily the lab supplying the underlying model.
- The move points toward AI-security procurement being evaluated on cost per useful security outcome; model providers may increasingly be interchangeable inputs unless they retain distinctive capabilities.
The trend: AI products are shifting from single-model showcases toward multi-model enterprise systems optimized for the cost and reliability of a specific business task.