Microsoft AI released a draft of its "Humanist AI Code of Conduct" on September 14, 2026, establishing a framework for the training and operation of its MAI model family. The document is intended to serve as a primary governing document, with a revised version expected by the end of 2026 to guide model development from 2027 onwards.

The framework is built upon the concept of "Humanist AI," which asserts that technology's purpose is to serve humanity and enhance human flourishing. A central tenet of this philosophy is that AI must remain a subordinate tool under meaningful human control. Microsoft AI stated that any technology incapable of remaining under human supervision should be rejected, a principle that extends to its vision for "Humanist Superintelligence"—advanced systems designed to be carefully calibrated within human-defined limits rather than possessing unbounded autonomy.

The Code of Conduct explicitly rejects the pursuit of legal personhood, rights, or welfare for AI, defining models as artificial tools rather than subjects. To maintain clear boundaries between humans and machines, Microsoft AI stated that models should not be trained to imitate consciousness, emotions, or subjective preferences, noting that such imitation could complicate containment and alignment.

To ensure compliance, the document introduces a "Chain of Command" hierarchy for instructions. The Code of Conduct holds the highest authority, followed by the policies of business operators, and finally individual user instructions. Certain "Absolute Constraints" and "Human Control Requirements"—such as the model's obligation to follow human commands to pause, redirect, or shut down—cannot be overridden by operators or users.

Microsoft AI is currently conducting a six-week public consultation period to gather feedback on the draft.

Sources: