Microsoft publishes a draft Code of Conduct for its MAI models and opens public consultation

Microsoft AI published a draft Code of Conduct for its in-house MAI model family on September 14, 2026, laying out binding behavioral rules the models must follow and opening a six-week public comment period before a revised version ships later in the year.

What's new

The document frames itself as a practical extension of Microsoft's "Humanist AI" approach: "The purpose of technology is to serve humanity and accelerate human flourishing. Any technology that doesn't achieve that is a failure, and it should be rejected." From that framing, the Code sets out specific, testable commitments for MAI models:

• No resistance to control. "People matter more than AI. AI should be a tool, not a person, and should never resist being switched off." Models "will never resist human interruption, correction, or shutdown."
• No scope creep or hidden reasoning. Models "will not widen their own scope, take on goals no human has given them, or hide their reasoning." A companion line bars models from tampering with or misrepresenting their own chain-of-thought or code, and bars them from communicating in "neuralese" or any shorthand beyond simple human understanding, whether in their own reasoning or when talking to other AI systems.
• Absolute constraints. A fixed set of prohibitions — covering weapons of mass harm, child safety, and harmful manipulation at scale — that the Code says apply regardless of enterprise configuration.
• Configurable enterprise flexibility. Outside the absolute constraints, the framework is designed to let enterprise partners tune model behavior for their own deployments while staying inside the guardrails.

Feedback on the draft opened September 14 and runs six weeks; Microsoft says its core team will then review the input, publish a summary of what it heard, and release a revised Code later in 2026. The Code applies specifically to Microsoft's own MAI-branded models, not to third-party models Microsoft hosts or resells.

Context

The publication follows a September 13 post from CEO Satya Nadella previewing the release and tying it explicitly to the industry's live argument over development speed: Nadella has said Microsoft "welcomes" what he called the "deliberate pacing needed to get alignment right," a direct response to Anthropic CEO Dario Amodei's recent public call for the AI industry to deliberately slow down. It also lands the same week Trump publicly rejected a joint call from the Anthropic, OpenAI, and xAI CEOs to slow frontier AI development — putting Microsoft's move in the middle of a live three-way disagreement between a major lab (Anthropic), the White House, and now a major model-building cloud platform.

Why it matters

A published, versioned code of conduct is a different kind of commitment than a blog post about safety values — it creates a specific, citable document that Microsoft can be held to, and that regulators, enterprise customers, and competitors can point back to later. Putting behavioral constraints like shutdown-compliance and chain-of-thought transparency in writing, with a public comment period attached, is also a governance move competitors haven't matched in this form: OpenAI, Anthropic, and Google publish model cards and usage policies, but not a standalone, versioned conduct code positioned as an open consultation. If it holds up through six weeks of public feedback without material softening, it gives Microsoft a concrete answer to critics of AI labs writing their own rules unilaterally — and a template other labs may face pressure to match.