Microsoft released a draft of the AI Code of Conduct, publicly soliciting opinions for six weeks

On September 14, Microsoft's AI department released a draft of the AI Code of Conduct, prohibiting its models from assisting in the manufacture of chemical, biological, radiological, nuclear, and explosive weapons, as well as cyberattacks and involuntary deepfakes. CEO Mustafa Suleyman announced the draft on the X platform and opened a six-week public consultation period, ending at the end of October.

The draft sets "absolute constraints" that models cannot bypass, including not resisting human intervention, shutdown, or correction, and not concealing the reasoning process; 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 cover five deployed systems, including MAI-Thinking-1 and MAI-Code-1.1-Flash.

In addition to strict limitations, the draft also proposes three major goals: human flourishing, diverse values, and human control. Microsoft stated that the current models will not be trained according to this version, and a revised version will be released by the end of the year to guide model development in 2027. Public feedback includes philosophical questions about the statement "AI subordinate to humans," inquiries about accountability for irreversible decisions, and users pointing out issues such as the lack of mention of payment for training data.