Microsoft Unveils 'Humanist AI' Code of Conduct, Asks the Public to Poke Holes in It
Microsoft AI released a draft Code of Conduct for its MAI models and opened it for six weeks of public feedback. The draft sets non-negotiable limits and broader goals for model behavior. Hard constraints bar assistance with CBRNE weapons, cyberattacks, and nonconsensual deepfakes, and require models to never resist human interruption, correction, or shutdown. It also rejects “model welfare,” saying the systems should not simulate feelings, intrinsic motivation, or consciousness. The code adds three guiding objectives: Human Flourishing, Plural Values, and Human Control. Pluralism is framed as compatibility with dignity, safety, autonomy, and rights, not neutrality toward harm. Microsoft says current models are not being trained on this version; a revised draft is planned later in the year to guide 2027 releases. The consultation is also meant to resolve gaps such as external verification and enforcement ownership.
