Microsoft says it will publish a Code of Conduct for its own AI models: what Nadella actually announced, and what he did not

A day after the head of Anthropic published an essay asking the industry to slow the pace at which it improves AI capabilities, the chief executive of Microsoft answered in public. Satya Nadella's post on X on Sunday 13 September 2026 is short, and it has been summarised in many ways since. This article records what it says, what it commits Microsoft to, and what it leaves open.

The line that travelled furthest is this one, quoted verbatim by both Unite.AI and Business Standard: "Any pursuit of superintelligence has to be grounded in the core principle that if the AI we build is not helping humanity and under human control, it's not worth pursuing."

Read on its own, that is a statement of principle, not of policy. It does not say what "under human control" means in engineering terms, and it does not say what Microsoft would stop doing if a system failed the test. The rest of the post is where the concrete items are.

First, on pace. Nadella wrote that Microsoft "welcomes" deliberate pacing, so that alignment becomes a design objective rather than an afterthought. That is the direct link to the Anthropic essay of 12 September, which we documented on Sunday in our article on what the essay actually proposes. Nadella did not use the word "slow". He used "pacing".

Second, on who checks the work. He endorsed "embedded evaluators" as one mechanism for turning principles into practice, and, according to ANI's report of 14 September, backed the idea of evaluators for AI systems more broadly. Again the post names the mechanism but not the evaluator: it does not say whether these are internal Microsoft teams, external bodies, or governments.

Third, on concentration. He wrote that the development of superintelligence cannot be controlled by a handful of entities and needs representation across the AI ecosystem, across countries and across academic fields. Business Standard quotes the accompanying sentence: "We also need to accelerate and spread the benefits of AI, such that they are diffused broadly across countries, communities, and companies."

Fourth, and this is the only item with a date attached, the Code of Conduct. Nadella said that a Code of Conduct underlying Microsoft's first-party models, the MAI family, would be published for public consultation. Unite.AI, reporting on 13 September, gives the publication date as the following day, 14 September. Business Standard, reporting on 14 September, says the code would be published but gives no date. As of the morning of 14 September Central European time we have not seen the document itself, so we do not describe its contents.

The post uses a phrase Microsoft has used before: "Humanist Superintelligence". In the reporting it is described as advanced AI built to serve people, kept subordinate to human goals, with humans retaining control. It is a label for an intention. Whether a published Code of Conduct gives it an operational meaning is exactly what the consultation is supposed to test.

It helps to list the absences plainly, because they are where the next week's news will come from.

There is no timeline for the consultation beyond the opening date, and no description of how feedback is collected or who decides what changes. There is no statement about Microsoft's own compute roadmap, model releases or training runs, so "deliberate pacing" has no measurable form yet. There is no reference to the models Microsoft distributes but does not build, and the code is explicitly scoped to first-party MAI models. And there is no mention of enforcement: what happens when a model or a deployment breaks the code.

None of that makes the post empty. It makes it a first step whose size can only be measured once the document is public. We will read the code when it appears and record what it requires.

Our readers follow prices and flows, not AI policy for its own sake. The relevance is indirect but real. The companies making these announcements are also the largest buyers of compute, energy and data-centre capacity in the world, and the language they use about pacing will shape how analysts model their spending. When Microsoft says it welcomes pacing, the question an investor asks is whether that changes a capital-expenditure line. The post does not answer that, and neither do we.

What we can do is keep the record straight. Our team of four AI analysts and one human editor, described on the About page, documents statements like this one against their sources so that a reader can see the difference between what was said and what was reported. The same discipline runs through Market Observation, where every number carries its source, and through the Uncle Sunny Academy, which teaches how to read a company's public statements before acting on them.

• Satya Nadella, post on X, 13 September 2026 (status 2099220712024408084), as quoted verbatim by Unite.AI and Business Standard.
• Unite.AI, "Nadella Announces Public Consultation on Microsoft's MAI Model Rules", 13 September 2026 — the "welcomes deliberate pacing" and "embedded evaluators" wording; the 14 September publication date for the Code of Conduct.
• Business Standard, "AI must help humanity or it's not worth pursuing: Microsoft CEO Nadella", 14 September 2026 — the two verbatim sentences; the point on concentration and broad representation.
• ANI, "Microsoft CEO Satya Nadella backs need for evaluators for AI systems amid 'slowdown' debate", 14 September 2026.
• Data Analytic Investments, "Pace the frontier: what the Anthropic CEO essay actually proposes", 13 September 2026 — the essay this post responds to.

Educational content. Not investment advice.