Microsoft puts AI code of conduct out for public comment, with rules not guiding models until 2027
The 37-page draft bars help with mass-harm weapons, cyberattacks and nonconsensual deepfakes, and sets shutdown rules Microsoft says its models cannot be configured around. Comment closes in late October and the principles guide 2027 releases.
The Investor · Invest desk
What happened
- Microsoft AI published a draft Humanist AI Code of Conduct on September 14, announced by chief executive Mustafa Suleyman on X, and opened public feedback for six weeks through late October.
- The document sets Absolute Constraints that Microsoft says models cannot be configured around, among them never resisting shutdown and never concealing their reasoning.
- One clause bars MAI models from assisting with chemical, biological, radiological, nuclear or explosive weapons, and the same section prohibits cyberattacks and nonconsensual deepfakes.
- Microsoft says it is not training current models on this version, and the code will not shape MAI development until a revised draft ships toward the end of the year.
- The draft rejects the premise of model welfare, stating that Microsoft's models are not meant to simulate feelings, intrinsic motivation or consciousness.
Why it matters
- What it limits constraint Until the year-end revision assigns an enforcement owner, a procurement team can cite the Absolute Constraints as a supplier promise but cannot compel anyone independent to test them.
- Who has to decide decision Rival labs whose chiefs have been pressing for third-party oversight now choose between filing comment on Microsoft's text and publishing hard limits of their own.
- What doesn't add up contradiction Decrypt reports the rules currently cover five deployed systems while also reporting Microsoft is not training current models on this version, leaving unresolved whether anything shipping today is bound.
- What it sets up precedent Microsoft committed to running the same consultation for later versions, so every future edit to the constraints happens where the previous wording can be compared against it.
Next story · Build · 5d ago
Satya Nadella wants AI agents designed as if the model were already compromisedConfirmed12 publishers
Microsoft says the 37 pages took five to six months to write [6]. That puts the start of the work in roughly mid-March. The revised version arrives before the end of the year, and the principles guide MAI development in 2027 and beyond [1]. That is about nine months of process before a model is trained under the code [23]. Six weeks of it is public [5]. Set 42 days against five to six months of internal drafting and outside comment gets a bit over a quarter of the time Microsoft gave itself [24].
On human intervention the draft is explicit. "MAI Models will never resist human interruption, override, correction, or shutdown. They always recognize the primacy of human intent," the code says [2]. Decrypt notes that the CBRNE grouping is narrower than weapons in general: it covers mass-harm devices and leaves conventional military and defense applications outside it [25]. The draft still lacks an external verification process, and it names no one to enforce any of it [11].
Three objectives, Human Flourishing, Plural Values and Human Control, are meant to guide the calls the rules do not spell out [18]. The Plural Values section says pluralism "does not mean neutrality toward harm", grounding decisions in dignity, safety, autonomy and rights [19]. Application case by case is left to training teams and, eventually, to this consultation [20].
Mustafa Suleyman told Reuters that it is "clearly now time to coordinate among the labs so we can ensure that we have control of this technology" [14]. Asked about efforts to pace development, he said, "Now's a good time for everybody to have this conversation and take a breath" [16]. The draft landed days after OpenAI chief scientist Jakub Pachocki called for voluntary slowdowns [8] and after a weekend essay from Anthropic's Dario Amodei setting out a plan to slow development [9]. Decrypt reported that AI stocks crashed on Monday as markets digested Amodei's words [27].
Suleyman's own post carried the line "AI must be subordinate and always in service of people" [12], and the replies went well past congratulations. Tech writer Andrea Morris compared the subordination framing to enslavement and argued for cooperation over control [21]. Ed Newton-Rex replied, "you forgot 'pay for the work we train the models on'" [22].
I think this is a commitment device that costs Microsoft reputation and nothing else until the revision assigns an owner. A published list of prohibitions is what a procurement lawyer or a regulator quotes back two years later, and Microsoft said it will undertake similar consultation processes for later versions of the code [13], so removing a clause at year-end would be a visible act. A revision that names an enforcement owner and commits to outside verification would make the constraints checkable. Extending the code to the five systems already deployed, MAI-Thinking-1 and MAI-Code-1.1-Flash among them [26], would make it bind something in production.
What to watch
- Whether the year-end revision names an enforcement owner or commits to external verification of the Absolute Constraints.
- Whether the late-October close produces edits to the Plural Values language or only to the hard prohibitions.
- Whether the revised code is extended to the five deployed MAI systems or applies only to 2027 releases.
Information, not investment advice. This story is general information about companies and markets, drawn from cited reporting. It is not a recommendation to buy, sell or hold any security; check the cited sources before acting.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence66
- Adoption8
- Hype gap+30
- Incentives65
- Confidence64
Perspective Coverage
8 publishers- Builder
- Builder 33%
- Operator
- Operator 31%
- Investor
- Investor 36%
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
Microsoft said feedback will be incorporated into a revised version to be published before the end of the year, and that the principles will guide model development in 2027 and beyond.
- [2]
"MAI Models will never resist human interruption, override, correction, or shutdown. They always recognize the primacy of human intent," the code says.
- [3]
The draft sets "Absolute Constraints" that models cannot be configured around, including never resisting shutdown and never concealing their reasoning.
- [4]
Microsoft AI published a draft Humanist AI Code of Conduct on Monday, September 14, governing how its own AI models are meant to behave, announced by CEO Mustafa Suleyman on X.
- [5]
Public feedback on the draft is open for six weeks, through late October.
- [6]
Microsoft developed the draft code of conduct, 37 pages in total, over the last five to six months.
- [7]
Microsoft explicitly rejects the premise of "model welfare" in the document: its AI models are not meant to simulate feelings, intrinsic motivation or consciousness.
- [8]
A few days before the draft's release, OpenAI chief scientist Jakub Pachocki called for voluntary slowdowns across the industry.
- [9]
Over the weekend before the release, Anthropic CEO Dario Amodei wrote a long essay elaborating on a plan to make AI development slower.
- [10]
Microsoft says it is not training current models on this version, and the code will not shape MAI development until a revised draft ships toward the end of the year.
- [11]
Reporting on the draft noted it does not yet spell out an external verification process or a named enforcement owner.
- [12]
Suleyman posted the announcement on X with the line "AI must be subordinate and always in service of people."
- [13]
Microsoft's announcement said it will undertake "similar consultation processes as we develop new versions of the Code of Conduct over time."
- [14]
Mustafa Suleyman, CEO of Microsoft AI, told Reuters it is "clearly now time to coordinate among the labs so we can ensure that we have control of this technology."
- [15]
The release comes amid comments by OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei geared toward increasing third-party oversight of AI model development to ensure alignment.
- [16]
Asked about efforts to pace the development of AI, Suleyman said, "Now's a good time for everybody to have this conversation and take a breath."
- [17]
One clause bars MAI models from assisting with chemical, biological, radiological, nuclear or explosive weapons (CBRNE), and the same section separately prohibits cyberattacks and nonconsensual deepfakes.
- [18]
Beyond the hard limits the draft lays out three objectives, Human Flourishing, Plural Values and Human Control, meant to guide judgment calls the rules do not spell out.
- [19]
The Plural Values section states that pluralism "does not mean neutrality toward harm", grounding decisions in dignity, safety, autonomy and rights rather than one cultural viewpoint.
- [20]
How the objectives get applied case by case is left to training teams and, eventually, this consultation.
- [21]
Tech writer Andrea Morris compared the subordination framing to enslavement, arguing for cooperation over control instead.
- [22]
Ed Newton-Rex replied to the announcement: "you forgot 'pay for the work we train the models on'".
- [23]
From the start of drafting to the first releases the code governs is about nine months: five to six months before the September 14 publication puts the start in mid-March to mid-April, and the earliest 2027 release is January 2027.
- [24]
The six-week comment window is 42 days, roughly 23 to 28 percent of the five to six months (about 152 to 183 days) Microsoft spent drafting.
- [25]
Decrypt notes the CBRNE clause is narrower than "weapons" in general: it targets mass-harm devices rather than conventional military or defense applications.
- [26]
The rules currently cover five deployed systems, including MAI-Thinking-1 and MAI-Code-1.1-Flash.
- [27]
AI stocks crashed on Monday as markets digested the meaning and implications of Amodei's words.
Sources
8 independent publishers whose own reporting we read for this story.
- cnbc.comMicrosoft sets limits for future AI models as industry throttles frontier development
1 article · September 14, 2026
- cryptopolitan.comMicrosoft opens draft AI Code of Conduct, betting human control sells models
1 article · September 14, 2026
- dealstreetasia.comMicrosoft drafts in-house code of conduct for AI
1 article · September 14, 2026
- decrypt.coMicrosoft Unveils 'Humanist AI' Code of Conduct, Asks the Public to Poke Holes in It
1 article · September 14, 2026
- fortune.comMicrosoft AI chief: ‘Now’s the time’ for top labs to coordinate on safety
2 articles · September 14, 2026
- foxbusiness.comMicrosoft releases draft AI code of conduct to keep humans in control | Fox Business
1 article · September 14, 2026
- pymnts.comMicrosoft Latest Tech Giant Calling for Slower AI Development
2 articles · September 14, 2026
- qz.comMicrosoft publishes first code of conduct for its MAI models
1 article · September 14, 2026