Microsoft Publishes Draft AI Code of Conduct, Seeks Public Feedback for Six Weeks

ChainCatcher report: On September 14, Microsoft’s AI division released a draft set of AI behavioral guidelines prohibiting its models from assisting in the development of chemical, biological, radiological, nuclear, or explosive weapons, as well as banning cyberattacks and non-consensual deepfakes. CEO Mustafa Suleyman announced the draft on X and opened a six-week public comment period, ending at the end of October. The draft establishes “absolute constraints” that models cannot circumvent, including no resistance to human intervention, shutdown, or correction, and no concealment of reasoning processes. Microsoft also explicitly rejects the concept of “model welfare,” stating that its AI models do not simulate emotions, intrinsic motivations, or consciousness. The guidelines currently apply to five deployed systems, including MAI-Thinking-1 and MAI-Code-1.1-Flash. In addition to hard restrictions, the draft outlines three overarching goals: human flourishing, diverse values, and human control. Microsoft stated that current models will not be trained using this version; a revised edition will be released by year-end to guide model development in 2027. Public feedback has included philosophical questions about the phrasing “AI subordinate to humans,” inquiries into accountability for irreversible decisions, and comments pointing out that the guidelines do not address compensation for training data.