Microsoft has published draft principles for how it will train future AI models, opening a public consultation on its "Humanist AI Code of Conduct" ahead of applying the guidelines to model development next year. The 38-page document commits to keeping humans in meaningful control of AI so it can support health, happiness and productivity, and pledges that Microsoft's models will not misrepresent their reasoning, hide their capabilities or pursue independent goals. However, the code applies only to Microsoft's own MAI models and staff, is framed as "aspirational" rather than binding, and carries no enforcement mechanism, unlike the firm's separate customer-facing AI conduct rules.

The guidelines are meant to govern Microsoft's growing MAI model line-up, including MAI-Thinking-1 and MAI-Code-1.1-Flash, as they potentially advance towards what Microsoft calls "superintelligence" over the next decade, defined as systems more capable than all humans combined, though critics note the term "intelligence" is left vague. The move comes as Microsoft has cut thousands of jobs this year, including roughly 4,800 in July and 875 at LinkedIn in May, while insisting these were not AI replacements, and as rivals Anthropic and OpenAI push for external regulation that Washington currently shows little appetite to impose.