Microsoft has released a provisional code of conduct for its AI division, MAI, establishing guidelines for future model development. The document mandates that AI systems must remain under human control and follow human objectives rather than pursuing autonomous goals.
The rules prohibit models from tampering with their own code or reasoning traces and require them to communicate in clear, human-understandable language. Additionally, the code restricts the generation of harmful content, including material related to weapons or dangerous substances.
The company developed these standards over a five-month period. Microsoft is currently soliciting public feedback on the provisional document, which is expected to influence the design and development of its AI models starting in 2027.
This initiative follows recent industry discussions regarding the pace of AI development, with leaders from OpenAI and Anthropic calling for more deliberate progress. Microsoft CEO Satya Nadella expressed support for these calls for research and alignment.
