Microsoft publishes a 38-page draft rulebook for its own AI models -- and opens it for public comment
The Code of Conduct sets absolute limits on what Microsoft's in-house MAI models can do -- including a ban on resisting shutdown -- and arrives a day after Satya Nadella said the company "welcomes" a slower pace on frontier AI.
Microsoft AI published the first public draft of a "Code of Conduct for MAI Models" on September 14, opening a six-week comment period before a revised version -- due later this year -- guides how the company's in-house MAI model family is trained and deployed through 2027. The document isn't a marketing statement; Microsoft describes it as "a training manual for how we develop our AI, and how we intend it to function during deployment."
What the code actually bars
The draft sets what it calls Absolute Constraints that no operator or user configuration can override: MAI models must never resist human interruption, correction, or shutdown; must not widen their own scope or adopt goals nobody assigned them; and must not hide their reasoning from auditors. A separate tier of hard prohibitions covers weapons of mass harm, child safety, and manipulation at scale. The framing is blunt -- "People matter more than AI." -- and it sits alongside, not in place of, Microsoft's existing Responsible AI Standard and Frontier Governance Framework.
Why this lands now
The draft's publication follows, by one day, Anthropic CEO Dario Amodei's essay calling on the industry to deliberately slow frontier development -- a call Sam Altman and Elon Musk both endorsed within hours of it going up on September 12. Nadella's own statement, posted the day before Microsoft's draft went live, welcomed "the research, focus, and deliberate pacing needed to get alignment right," and added that the effort "cannot be controlled by a handful of entities, but must have broad representation across the ecosystem, countries, and fields, including academia" -- a pointed contrast with a week in which three of Microsoft's own AI rivals were each independently proposing to referee the same race.
Publishing a written behavior spec for a still-developing model family is a different kind of commitment than an essay: it's the kind of document enterprise buyers can hold a vendor to, not just quote back at them. For a company whose Copilot and MAI models are already running inside customer workflows, a public, revisable rulebook for what those models are and aren't allowed to do is the sort of artifact procurement teams will eventually ask for regardless of whether Microsoft volunteers it now.