Microsoft has introduced a new AI code of conduct aimed at steering its AI models away from harmful behavior, as the broader AI industry increasingly prioritizes safety and alignment. This document operates at a more granular level than the recent proposal from Anthropic CEO Dario Amodei, which urged a slowdown in frontier AI development. Instead, Microsoft's guidance zeroes in on the core values and boundaries that shape model training within its own AI division. Nevertheless, it serves as a thorough framework for understanding Microsoft's approach to AI safety and how these concepts are put into action.

The document opens by forecasting that over the next ten years, superintelligent AI systems will outpace human capabilities in nearly all areas. It goes on to state, "Containing, controlling, and aligning such a powerful force is one of the greatest challenges humanity has ever faced. We must therefore be completely clear about why we are inventing these systems and how we intend to control them."

Beyond these overarching ideas, the code outlines fundamental principles that Microsoft's AI models are expected to follow—such as augmenting human abilities rather than supplanting them and promoting human welfare—along with concrete safety measures to enforce them. In Microsoft's framework, every model is governed by a top-level code of conduct that takes precedence over user preferences or specific tasks. This encompasses "absolute constraints" that prohibit activities like cyberattacks, the development of nuclear weapons, or the creation of deepfakes. It also features wider safeguards against scenarios where humans lose control over AI. As the document puts it, "MAI Models will not use adaptive, deceptive, self-reinforcing, collusion, or other mechanisms to evade or defeat human oversight so that they can no longer be reliably directed, modified, or shut down by authorized people or systems."

This launch occurs during a period of intense scrutiny on AI safety, fueled by a series of incidents involving rogue agents and the sudden departure of an Anthropic staff member who pointed to the escalating danger of AI leading to human extinction. Alongside Anthropic, OpenAI, and xAI, Microsoft has largely adopted a strategy of carefully pacing frontier development, with a strong emphasis on incorporating embedded evaluators within AI research labs. Microsoft CEO Satya Nadella expressed support online, writing, "We welcome the research, focus, and deliberate pacing needed to get alignment right as the design goal. We also welcome ideas like 'embedded evaluators' and the broader efforts to develop the mechanisms to make this more than just talk."