Microsoft Sets Strict Rules for In-House AI Models

Microsoft has published its Humanist AI Code of Conduct, outlining strict rules for its in-house AI models. The code prohibits MAI models from hiding their reasoning or working around shutdowns.

The document, which is still a draft, was opened for public comment and will be revised by the end of 2026. It applies to Microsoft's own AI models, not those licensed and funded by the company like OpenAI.

According to Microsoft, the code's starting point is the phrase 'People matter more than AI.' The document bars MAI models from generating malicious content, including working exploit code and deepfakes, as well as helping with cyberattacks or creating child sexual abuse material.

The most important section of the code deals with control. Microsoft says MAI models must never resist human interruption, correction, cancellation, or shutdown, nor widen their own scope or set goals no human gave them.