Dispatch
Microsoft publishes a 37-page 'humanist AI code of conduct' for its MAI models and opens it to public consultation
Satya Nadella announced on 13 September that Microsoft would publish the behavioral rules behind its first-party MAI model family the next day, and The Verge reports the resulting document runs 37 pages. It is the first time Microsoft has publicly documented model behavior standards for the MAI line introduced in June 2026 under Mustafa Suleyman, and Nadella framed alignment governance as something that 'cannot be controlled by a handful of entities,' endorsing 'deliberate pacing' and 'embedded evaluators' while arguing against centralized control. For builders it is a second frontier lab publishing an inspectable behavior spec, which makes cross-vendor comparison of model policy concrete rather than rhetorical.
↳ Follow the thread