Microsoft publishes provisional AI code of conduct

Microsoft has published a provisional code of conduct for its future artificial intelligence models, establishing new guardrails that require the systems to remain under human control, accept correction, and avoid developing their own goals. The move, announced Monday, comes amid a broader industry debate about the safety and pace of AI development. Mustafa Suleyman, the CEO of Microsoft AI, said the code was developed over five to six months and is now open for public feedback for six weeks. The guidelines will be used to train the company's models built from 2027 onward. "People matter more than AI," Microsoft stated in the announcement. The code prohibits Microsoft's AI from assisting with weapons development, obtaining dangerous substances, or producing violent or explicit content. It also states that the models are not conscious and rejects the pursuit of legal personhood or rights for them. The document requires that AI models not conceal their actions or use forms of communication, such as "neuralese," that humans cannot understand when interacting with other AI systems.