Microsoft has introduced a draft human-centric code for its MAI division models, mandating that AI systems remain under human control. The document stipulates that artificial intelligence must not mimic consciousness, feelings, or personal preferences, nor should it possess legal personhood or rights.
The proposed rules require that AI models remain auxiliary tools that cannot resist human intervention, such as shutdown or redirection. Furthermore, AI agents are prohibited from using internal machine languages that are incomprehensible to humans or concealing their reasoning processes.
The company has initiated a six-week public consultation period to refine the code, which is intended to guide model development starting in 2027. These requirements apply exclusively to Microsoft AI systems and do not govern third-party models hosted on the company's infrastructure.
The policy also addresses user interaction, banning models from fostering emotional dependency or engaging in excessive flattery. Microsoft emphasizes that accuracy and utility must take precedence over providing answers that merely align with user expectations.
This initiative follows broader industry discussions regarding the pace of AI development. Microsoft leadership has expressed support for independent testing and has indicated a willingness to prioritize safety over the pursuit of autonomous systems that could bypass established constraints.
