Microsoft AI puts MAI model code of conduct up for discussion

🛡 Microsoft AI puts MAI model code of conduct up for discussion

On September 14, Microsoft AI published a draft of the "Humanist AI Code of Conduct" for MAI models. The discussion is open for six weeks, the final version is due by the end of the year, and it will be used in model training starting in 2027. The code requires not resisting interruption and shutdown, not setting goals without a human request, and not hiding reasoning from auditors.

🌍 For the first time at a major company, the rules for frontier models have been formalized as a public regulatory document: unbreakable Absolute Constraints are above operator policies, and operator policies are above user requests. A model fails a task if its completion violates the code.

👤 The full text is available at microsoft.ai/code-of-conduct, and feedback is accepted via the Give Feedback form until the end of the six-week window. Section 2.2 "Chain of Command" and the MAI-Thinking-1 response examples are useful.

Source 1: https://microsoft.ai/news/mai-code-of-conduct/