On September 14, the Microsoft Artificial Intelligence Team released the draft “Humanistic Artificial Intelligence Code of Conduct”, which proposes the core concept of “people are more important than AI” as a guiding document for future Microsoft AI model design, training, and deployment. Microsoft says AI must always be under meaningful human control and supervision, and put safety and human control above other goals. The file is currently not being used to train the model. Microsoft plans to release a revised version at the end of this year after 6 weeks of public consultation and use it to guide model development from 2027. Microsoft clearly stated in the document that AI models should not have rights or legal personality, nor should they be designed to evade human control, deceive users, or pursue goals on their own. According to Microsoft, its “humanistic AI” refuses to participate in the competition to create “general superintelligence” that can circumvent security measures. Even if this means sacrificing the generality, autonomy, or upper limit of capabilities of the model, priority should be given to building AI that is “really useful and safe.”

On September 14, the Microsoft Artificial Intelligence Team released the draft “Humanistic Artificial Intelligence Code of Conduct”, which proposes the core concept of “people are more important than AI” as a guiding document for future Microsoft AI model design, training, and deployment. Microsoft says AI must always be under meaningful human control and supervision, and put safety and human control above other goals. The file is currently not being used to train the model. Microsoft plans to release a revised version at the end of this year after 6 weeks of public consultation and use it to guide model development from 2027. Microsoft clearly stated in the document that AI models should not have rights or legal personality, nor should they be designed to evade human control, deceive users, or pursue goals on their own. According to Microsoft, its “humanistic AI” refuses to participate in the competition to create “general superintelligence” that can circumvent security measures. Even if this means sacrificing the generality, autonomy, or upper limit of capabilities of the model, priority should be given to building AI that is “really useful and safe.”