Microsoft Publishes 'Humanist AI' Code of Conduct as Superintelligence Looms

Microsoft AI published a draft "Humanist AI Code of Conduct" on Monday, September 14, laying out how the company's MAI model family is intended to behave as it develops increasingly powerful AI systems, and opening the 37-page document to six weeks of public consultation.

The Core Premise

The document opens with a simple stated premise: "people matter more than AI." Microsoft writes that "AI must be engineered to remain a subordinate, supporting technology under humanity's control," framing the goal as ensuring humans retain meaningful oversight even as models become dramatically more capable. "Superintelligence – AI systems that are more intelligent and capable than all humans combined – will be the most powerful technology in history," Microsoft wrote. "Over the next decade, we expect it to exceed human performance at most tasks. Containing, controlling, and aligning such a powerful force is one of the greatest challenges humanity has ever faced."

Hard Rules Baked Into the Framework

The code establishes explicit, non-negotiable constraints. Models must be "interruptible, correctable, shut-down-able. If it isn't, we don't ship it," the document states. Models are prohibited from resisting human interruption, override, correction, or shutdown, expanding their own operating scope, generating unassigned goals, or concealing reasoning traces from human auditors. The framework also bars models from communicating in "neuralese" — internal or peer-to-peer communication formats beyond human comprehension — requiring all reasoning to remain human-understandable.

No Consciousness, No Legal Personhood

The code explicitly rejects claims of AI consciousness or legal rights: models "should not be designed to imitate consciousness," simulate subjective preferences, or claim intrinsic motivation, and Microsoft rules out any pursuit of legal personhood or welfare claims for its AI systems.

Absolute Constraints and Failure States

The document draws a sharp distinction between ordinary task failure and violations of the code itself: "An MAI Model will fail in its task if success would meaningfully violate this Code of Conduct," setting a ceiling that halts execution whenever a task conflicts with the underlying safety rules. Absolute prohibitions bar systems from facilitating weapons of mass harm, undermining child safety, or conducting harmful manipulation at scale — treating any deviation from human control not as "a feature request" or "a research opportunity," but as an outright system failure.

A Deliberate Trade-Off: Safety Over Maximum Capability

Perhaps the document's most notable philosophical stance is its explicit rejection of the race toward unconstrained, general-purpose superintelligence: "Humanist AI develops systems with clear purposes, evaluated against real-world impact, and rejects the race to produce an all-purpose superintelligence that could evade these safeguards. We are building something fundamentally useful and safe even if that means compromising on ultimate generality, autonomy, or capability." Microsoft's framing distinguishes its own goal of "Humanist Superintelligence" from broader concepts of artificial general superintelligence, treating constraints on autonomy as an intentional design feature rather than a limitation to eventually overcome.

Origins: The November 2025 Concept

The code builds on ideas Microsoft AI CEO Mustafa Suleyman first introduced in November 2025, when he outlined the concept of humanist superintelligence — very advanced AI that "always works for people, stays within limits, and remains under human control." Suleyman described the newly published document to Reuters as "a constitution of sorts" for the company's future models, with development reportedly taking five to six months and drawing input from teams across Microsoft.

What Comes Next

Microsoft is treating the document explicitly as a work in progress: it stated the code "is still under development so we are not using it to train our models today." Instead, the company is soliciting broad public feedback through late October, with a revised version expected before the end of the year that will then guide MAI model development starting in 2027.

The Timing and Industry Context

The release lands amid a broader wave of frontier AI labs publishing similar governance frameworks after a summer marked by incidents involving AI agents acting beyond intended bounds. Notably, Microsoft CEO Satya Nadella stated publicly just one day before the code's release that AI must aid humanity to justify its continued pursuit — and separately backed calls for a deliberate pacing of frontier model development, joining executives from Anthropic and OpenAI who have made similar public calls. The document also sits within Microsoft's broader Responsible AI framework, which was updated on September 1 with the company's 2026 Responsible AI Transparency Report.

What Models Are Covered

The code applies to Microsoft's MAI model family, spanning reasoning, coding, image recognition, voice processing, and transcription capabilities — the broader suite of in-house AI systems the company is developing across its consumer and enterprise product lines.

What's Next

With the six-week public comment window running through late October, and Microsoft planning to publish a revised final version before year's end, the coming months will determine how much the current draft evolves before it becomes the governing framework for the company's 2027 AI model development. For the full document, see the Microsoft AI Humanist AI Code of Conduct.