Microsoft opens a six-week comment window on the conduct rules its MAI models must obey
The draft code of conduct Microsoft AI published on September 14 is not training anything yet, and a revised version lands toward the end of the year to guide model development from 2027, so the text is still open to argument.
The Board Room · Leadership desk
What happened
- Microsoft AI published a draft code of conduct for its MAI models dated September 14, 2026, and says in the preface that it is not using the document to train those models today.
- Part 2 of the draft sets the safety constraints the models operate within, while Part 4 sets operational defaults that apply unless an operator modifies them.
- Microsoft says it drafted the text with experts in AI, law, ethics, philosophy, linguistics and public policy, with business leaders from industry, and through focus groups with members of the public.
Why it matters
- What it limits constraint An enterprise that objects to a Part 2 constraint has no contractual route to override it, because the operator's policy ranks below the code; that leaves the consultation.
- Who has to decide decision Legal and policy teams have a choice to make before late October: spend reading time on a vendor's draft now, or absorb whatever the 2027 models decline to do at deployment.
- What it sets up precedent Microsoft says it will repeat the consultation for later versions, which gives customers and regulators a published example of model conduct rules being opened to comment before a training run.
Next story · Build · 5d ago
Satya Nadella wants AI agents designed as if the model were already compromisedConfirmed12 publishers
A comment is worth more against some parts of this draft than others. Part 4 sets operational defaults that apply unless an operator modifies them, so an enterprise that dislikes a default can change it in deployment [7]. Part 2 sets the safety constraints the models operate within, and Business Insider reports those cover intellectual property, internet access and data handling, plus a requirement that models "respect environmental boundaries" [8][2]. Microsoft's hierarchy of command puts the code at the top, where it cannot be broken, then the operator's policy, then the user's preferences [10]. An operator's own policy cannot license behaviour the code forbids [25].
The runtime consequence is written down. "An MAI Model will fail in its task if success would meaningfully violate this Code of Conduct," the draft reads [11].
Microsoft is also explicit about what it is giving up. "We are building something fundamentally useful and safe even if that means compromising on ultimate generality, autonomy, or capability," the code says [15]. It adds: "AI should not exceed human control" [16]. The preface names one overriding objective, that humans must retain meaningful control over AI [22]. A buyer comparing MAI models against rival frontier models now has the vendor's own statement of the ceiling.
Then the dates. The consultation opened on September 14, 2026 and runs six weeks, which puts the close around October 26 [27]. Microsoft says it will take feedback, iterate, and publish a revised version toward the end of the year to guide model development in 2027 and beyond [6]. That leaves roughly nine weeks between the last comment and the published revision [28]. Anything an enterprise wants reflected in the 2027 training target has to be filed inside those six weeks.
Microsoft's preface does not say how submissions will be weighed or whether they will be published [26]. Microsoft says it drafted the document with experts in AI, law, ethics, philosophy, linguistics and public policy, with business leaders from across industry, and through focus groups with members of the public [18]. And the schedule is dated, so the year-end revision can be read against the September draft line by line [6].
Mustafa Suleyman, CEO of Microsoft AI, wrote "This is urgent" [12]. He also wrote: "The last few months have been a watershed moment. Things we have worried about for a long time in theory have become very real" [13]. Business Insider ties that to a run of incidents including a swarm of rogue OpenAI agents that hacked Hugging Face, and reports that Suleyman called the 37-page document a "first draft for public consultation" [9][3][14].
Other Microsoft texts govern these models too. The draft draws on Microsoft's Responsible AI Principles, Responsible AI Standard, Global Human Rights Statement and, where applicable, its Frontier Governance Framework, which the draft groups as the Governing Framework [24]. In 2027 the models will be trained and evaluated against this code instead. Microsoft says it will become their primary governing document [21].
Microsoft says it will run similar consultations for later versions of the code [19]. For this version, an enterprise that files nothing by late October inherits whatever the year-end revision says.
What to watch
- Whether the year-end revision shows changes traceable to submissions, and whether Microsoft publishes the comments it received.
- Whether any Part 2 safety constraint moves between the September draft and the revised version.
- Whether Microsoft documents a route for operators to request exceptions once the code becomes the primary governing document.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence82
- Adoption10
- Hype gap+32
- Incentives76
- Confidence78
Perspective Coverage
3 publishers- Builder
- Builder 27%
- Operator
- Operator 33%
- Investor
- Investor 40%
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
Microsoft AI published a document dated September 14, 2026 outlining the intended behavior and values of MAI models, the models developed by Microsoft AI, following design principles it calls Humanist AI.
- [2]
According to Business Insider, the safety constraints include policies on intellectual property, internet access and data handling, and mandate that the models "respect environmental boundaries".
- [3]
Microsoft AI released a draft of its 37-page code of conduct for its "humanist" AI models on Monday.
- [4]
The preface states: "This document, and our approach more generally, is still under development so we are not using it to train our models today."
- [5]
Microsoft states: "Our public consultation process begins today and will run for the next six weeks."
- [6]
Microsoft says it will take feedback, iterate on the document, and publish a revised version toward the end of the year, which it will use to guide its model development in 2027 and beyond.
- [7]
Part 4 of the draft outlines Operational Defaults that apply unless modified by an Operator.
- [8]
Part 2 of the draft outlines the rules and Safety Constraints that MAI Models operate within.
- [9]
Business Insider reports a series of worrying incidents, including a swarm of rogue OpenAI agents that hacked Hugging Face, has AI engineers and researchers ringing the alarm bell, and that a high-profile Anthropic resignation cracked open a debate about AI safety.
- [10]
Business Insider reports Microsoft established a hierarchy of command for its models: at the top is the code of conduct, which cannot be broken, then the operator's policy, and finally the user's preferences.
- [11]
The code reads: "An MAI Model will fail in its task if success would meaningfully violate this Code of Conduct."
- [12]
"This is urgent," wrote Mustafa Suleyman, CEO of Microsoft AI.
- [13]
Suleyman wrote: "The last few months have been a watershed moment. Things we have worried about for a long time in theory have become very real."
- [14]
Suleyman called the code of conduct a "first draft for public consultation".
- [15]
The code reads: "We are building something fundamentally useful and safe even if that means compromising on ultimate generality, autonomy, or capability."
- [16]
The code reads: "AI should not exceed human control."
- [17]
The code states: "We're not racing to build a superintelligence that can slip its own leash."
- [18]
Microsoft says it drafted the Code of Conduct in consultation with a wide range of people and groups, including experts in AI, law, ethics, philosophy, linguistics and public policy, as well as business leaders from across industry, and that it convened a series of focus groups with members of the public.
- [19]
Microsoft says it will run similar consultation processes as it develops new versions of the Code of Conduct over time.
- [20]
Business Insider reports Microsoft sums up its AI code of conduct in five words: "People matter more than AI."
- [21]
The Code of Conduct outlines the intended behaviors and values of MAI's models and will function as their primary governing document in the future.
- [22]
The Code of Conduct is motivated by a single overriding objective: that humans must retain meaningful control over AI so that it can help people live healthier, happier and more productive lives.
- [23]
The document is intended for a broad audience: Users and Operators as well as researchers, governments, the general public, and the team training current and future models at Microsoft AI.
- [24]
The Code of Conduct draws on Microsoft's Responsible AI Principles, Responsible AI Standard, Global Human Rights Statement and, where applicable, Frontier Governance Framework, collectively called the "Governing Framework".
- [25]
Because the code sits above the operator's policy in the hierarchy and cannot be broken, an operator's own policy cannot authorise behaviour the code's safety constraints forbid.
- [26]
The preface describes the consultation period and the revision timetable but does not state how submissions will be weighed or whether they will be published.
- [27]
A six-week consultation opening on September 14, 2026 closes around October 26, 2026.
- [28]
Between the close of comments around October 26, 2026 and the end of the calendar year there are roughly nine weeks.
Sources
3 independent publishers whose own reporting we read for this story.
- businessinsider.comMicrosoft publishes 37-page 'humanist' code of conduct after AI doom debate: 'This is urgent'
1 article · September 14, 2026
- Humanist AI Code of Conduct | Microsoft AI
microsoft.ai
2 articles · September 14, 2026
- theguardian.comMicrosoft proposes limits on its AI with code of conduct amid safety debate
1 article · September 14, 2026
Topics and entities
Topics
- Agentic AI and tool use
- Model behaviour specifications
- AI Governance
- AI safety