Microsoft has dropped a draft of its new human-first code of conduct for the MAI family of models. The document lays out the basics: AI must stay under human control, can't fake consciousness, and needs to stick to its assigned tasks. Microsoft is opening up the draft for public comment, saying it plans to tighten up the rules based on feedback.

What Microsoft Announced

The code sets the ground rules for building and using AI within the MAI lineup—putting human oversight front and center and banning any attempts at simulating consciousness. These limits are pitched as non-negotiable, forming the backbone for how models should behave and get used in real-world applications.

How It'll Roll Out

The document isn't being used to train neural nets just yet. Microsoft has kicked off a six-week public consultation, after which it plans to clarify and strengthen the language. The final version is set to become the foundation for MAI models starting in 2027.

Who the Code Covers

The rules apply to Microsoft AI division systems only. Third-party models aren't automatically covered by this code.