Microsoft's AI Code of Conduct: Human Control Takes Priority
Microsoft has released a draft code of conduct for its AI models, emphasizing human control over model capability and autonomy. The Humanist AI Code of Conduct prioritizes people over AI, stating that 'people matter more than AI.' The code is based on Microsoft's existing Responsible AI Principles and will serve as the primary governing document for its MAI model family.
The draft code outlines rules for model behavior, safety restrictions, instruction hierarchy, and autonomous actions. It establishes a clear hierarchy of control, with human oversight above model independence. If completing a task would violate the code's absolute safety constraints, the model is expected to fail rather than work around the restriction.
Microsoft's models must stay within the permissions and resources provided for a task and cannot independently expand their goals. They must also 'never resist human interruption, override, correction, or shutdown.' The company is seeking public feedback on the draft code for six weeks before revising and publishing another version by the end of the year.
The code will guide model development beginning in 2027 and applies only to models built by Microsoft AI. It does not cover outside models that Microsoft hosts or uses in its products, including those from OpenAI and Anthropic.