Microsoft Publishes Strict Rulebook for In-House AI Models

Microsoft has published its first AI behavior rulebook, a 37-page document called the Humanist AI Code of Conduct. The code outlines the rules for Microsoft's in-house artificial intelligence (AI) models, known as MAI, and aims to ensure that they prioritize human needs and safety above all else.

The code was published on September 14, 2026, and marks a significant shift in the company's approach to AI development. According to Unite.AI, Microsoft has spent years licensing OpenAI's models for its own products, but MAI is its own proprietary technology. The Humanist AI Code of Conduct is the first document that spells out how those models must behave.

One of the key principles outlined in the code is that human control and reliable safety should be the top priority when designing and deploying AI models. This means that Microsoft's AI models will not be able to cross certain boundaries, such as running cyberattacks or contributing to nuclear weapons development.