Microsoft introduced Project Perception, a cybersecurity platform that finds and fixes software vulnerabilities using models from Microsoft, OpenAI and Anthropic together.
What holds the product up is an orchestration layer deciding which model handles each task: a cheap model for triage, a frontier model only when the reasoning requires it. According to the company, that distribution cuts cost enough to make continuous scanning practical, which until now was too expensive to run without pause.
Choosing to include two direct competitors inside your own product is the part that says most about the state of the market. It admits publicly that no single vendor delivers the best cost-to-result ratio on every task, and that tying a product to one model is a commercial decision rather than a technical one.
The stated positioning is as a cheaper alternative to a specific Anthropic product line, which creates the unusual situation of selling against a company whose model the product uses.
For security teams, the practical point is the cost of continuous scanning. Analysis that runs once a week finds the problem a week later. The product's argument is that routing economics bring scanning closer to real time.