Microsoft published its AI rulebook as a draft and asked the internet to attack it
Microsoft made the unfinished format part of the message: invite criticism, name the weak spots, and promise to show what changed.
Microsoft AI did something unusual with a values document on September 14, 2026. Instead of publishing its Humanist AI Code of Conduct as a polished final statement, it published a draft, invited six weeks of public criticism, named questions it still thinks are unresolved, and promised to explain what changed afterward.
That makes the format as interesting as the policy. Companies regularly publish AI principles, ESG commitments and codes of conduct as finished declarations. Those documents ask audiences to judge the company's position. Microsoft's draft asks audiences to participate in shaping it, while creating a public test of whether the company later acts on what it hears.
Key Takeaways
- Microsoft turned a values document into a time-bounded consultation instead of presenting it as a finished corporate statement.
- The strongest credibility test is still ahead: Microsoft has promised to publish what it learned and what it changed.
- Communications teams can borrow the mechanics by setting a deadline, naming weak points and committing in advance to a visible revision record.
Table of contents
Jump to each section:
- What Microsoft actually published
- Four choices worth studying
- Plain language as positioning
- The limits of the format
- What to borrow
- The open question
What Microsoft actually published
The Microsoft AI announcement describes the Code of Conduct as a work in progress governing how its in-house MAI models are intended to behave. The premise is deliberately simple: people matter more than AI, AI should remain a tool rather than a person, and models should remain under meaningful human control.
The Code itself says the current draft is not being used to train Microsoft's models today. Microsoft says it plans to revise the document after consultation and use the resulting version to guide model development from 2027 onward.
The consultation runs for six weeks. Microsoft says the core drafting team will then review submissions, publish a summary of what it learned and changed, and release a revised version later in 2026, according to Microsoft AI.
The substance includes what Microsoft calls Absolute Constraints. These cover areas such as weapons of mass harm, child safety and harmful manipulation, alongside requirements that models remain subject to interruption, correction and shutdown.
Four choices worth studying
First, publish unfinished. A draft changes the audience's role. A finished statement invites judgment, while a draft invites participation and gives the company room to revise before every sentence becomes a permanent commitment.
Second, name your own weak points. Microsoft's feedback request asks where its language is too loose to evaluate and how it can be more concrete about ideas such as human flourishing. That is a notable communications choice because it turns an obvious line of criticism into an explicit assignment for readers.
Third, commit to showing the diff. Microsoft says it will publish what it learned and what it changed. That promise matters more than the invitation itself because it creates something outsiders can later verify. A consultation without a visible revision trail can become little more than an inbox.
Fourth, sequence the messengers. Microsoft CEO Satya Nadella previewed the broader human-control framing on X on September 13. Microsoft AI published the draft the next day. The sequencing is similar to the principle ContentGrip examined in Apple's CEO transition: different voices can carry different parts of one narrative without every message doing the same job.
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.— Satya Nadella (@satyanadella) September 13, 2026
Plain language as positioning
"People matter more than AI" is not a technical standard. It is a repeatable sentence placed at the front of a much more complex governance document, and it became the phrase most secondary coverage carried forward, including The Verge.
That is a useful reminder for communications teams working on dense policy material. The sentence that travels furthest is rarely the sentence with the most caveats. If a document needs one idea to survive headlines, executive posts and internal briefings, the plain-language line should be chosen deliberately rather than discovered after publication.
The limits of the format
A consultation is not the same thing as accountability. Independent governance analysis from Novara Consulting Group argues that the draft leaves important procedural questions unanswered, including audit rights, disclosure obligations and adjudication mechanisms for operators.
That does not invalidate the consultation, but it defines its limit. Microsoft wrote the document, Microsoft will review the comments, and Microsoft will decide what to incorporate. The company explicitly says it cannot promise that feedback will be adopted. The mechanism is therefore a test of transparency and responsiveness, not independent enforcement.
The timing also sits inside a wider industry debate about pacing advanced AI. Axios reported that Nadella backed calls for deliberate pacing before the draft appeared. That sequence is worth noting, but there is no need to claim the consultation was timed strategically unless Microsoft says so.
What to borrow
First, decide whether a draft serves you better than a declaration. If a communications or legal team already expects substantive criticism, publishing a controlled draft can create a defined channel for that criticism before the language hardens.
Second, set both a deadline and an output. Microsoft's model has a bounded consultation period and a promised revision summary. A generic "we welcome feedback" line has neither, so audiences cannot tell when consultation ends or what it produced.
Third, publish what changed. The accountable owner should be named before consultation begins, and the organization should commit to a change summary that links major outside concerns to the final wording. If the team cannot make that commitment, calling the process a consultation may raise expectations it cannot meet.
Fourth, choose the sentence non-specialists will repeat. Policy, legal and communications teams should agree on one plain-language line that accurately reflects the document without pretending to replace it. Microsoft's example shows how quickly that line can become the public shorthand.
Finally, decide how you will answer the "no teeth" critique. A voluntary code can still be useful, but teams should decide in advance whether they will point to contracts, audits, governance bodies or other enforcement mechanisms, or simply acknowledge that the document is a statement of intent.
The open question
Microsoft has created a clear test for itself. When the revised version appears later in 2026, readers will be able to compare the draft with the final text and ask whether outside input changed anything material.
That is what will determine whether the consultation was mainly a staging device or a genuine revision process. Publishing a draft turns critics into contributors. Publishing the changes is what makes that invitation verifiable.
