Microsoft Publishes Draft 'Humanist AI' Code of Conduct, Tells Its Models Never to Resist Shutdown
Microsoft Publishes Draft 'Humanist AI' Code of Conduct, Tells Its Models Never to Resist Shutdown
Microsoft AI has released a draft code of conduct, reportedly around 37 pages long, laying out rules for how its MAI models and any future superintelligence systems should behave. Published around September 14, 2026, the document frames artificial intelligence as a tool that must remain what Microsoft describes as "subordinate, aligned, and contained" relative to human authority. The guiding philosophy, as stated in the draft, is that "people matter more than AI."
The draft explicitly rejects granting AI systems legal personhood, welfare consideration, or rights. Microsoft has opened the document for a six-week public consultation period, with a revised version expected before the end of 2026.
The 'Absolute Constraints': Hard Limits on AI Behavior
Central to the draft is a set of what Microsoft calls Absolute Constraints — non-negotiable boundaries rather than adjustable guidelines. According to the document, these include:
- Bans on assisting with weapons of mass harm, including chemical, biological, radiological, nuclear, and explosive (CBRNE) threats
- Prohibitions on offensive cyber operations
- An explicit rule against evading human oversight or resisting shutdown commands
- Restrictions on large-scale manipulation of humans
Microsoft has framed these constraints as fixed limits intended to hold regardless of how capable future systems become.
Why Now: Context Behind the Announcement
The draft code follows Microsoft's formation of a dedicated superintelligence team in November 2025, under AI CEO Mustafa Suleyman. Suleyman has publicly emphasized that companies developing advanced AI must not "confuse control with cooperation."
Microsoft's own announcement also references what it describes as "large scale, highly coordinated, and persistent hacking campaigns of AI agents" as part of the motivation behind the new rules. This specific claim does not appear to be independently corroborated in secondary coverage reviewed, and it should be treated as an unverified assertion from Microsoft's own materials rather than an established fact.
How This Compares to Other AI Labs' Public Stances
Microsoft's move arrives amid a broader industry conversation about the pace and oversight of AI development. Anthropic CEO Dario Amodei has publicly called on the industry to "pace the frontier," and OpenAI CEO Sam Altman has been characterized as broadly agreeing with calls for a more cautious, safety-conscious approach. Other figures, including David Sacks, have reportedly pushed back against calls for mandated regulatory frameworks.
A recurring observation among commentators is that Microsoft's self-published, voluntary code stands in contrast to arguments for external, government-enforced regulation of AI development.
Open Questions and Skepticism About Enforceability
Several concerns have surfaced in commentary surrounding the draft. A recurring concern among consumers and industry watchers is whether concepts like "human flourishing" and phrases such as "meaningfully violate" the Code are defined concretely enough to be evaluated, audited, or enforced in practice.
It also remains unclear how the Absolute Constraints would be technically operationalized — for instance, how a model's internal reasoning could be reliably checked to confirm it is not resisting shutdown or expanding its own scope. Because the Code is authored and self-policed by Microsoft, some observers have raised questions about whether a company can meaningfully constrain its own AI ambitions without independent, third-party accountability.
There is also uncertainty about whether the six-week public consultation will meaningfully shape the final document, or largely serve as a formality ahead of a predetermined release. Separately, Microsoft has made a number of self-reported performance claims about specific MAI models — such as MAI-Transcribe-2 and MAI-Thinking-1 — that remain independently unverified and should be considered alongside that caveat.
What Happens Next
The six-week public feedback window is expected to close ahead of a revised release, with a finalized Code of Conduct anticipated before the end of 2026. How other AI companies respond, whether through similar voluntary frameworks or continued calls for external regulation, may shape the broader industry conversation around AI safety and oversight in the months ahead.