Microsoft AI published a draft "Humanist AI Code of Conduct" on Monday, a 38-page document laying out binding rules for the behavior of the company's in-house MAI model family — including a flat requirement that its systems never resist human correction, override or shutdown. The draft opens a six-week public comment period before Microsoft finalizes the code later this year.
The document reduces its position to five words, according to Microsoft AI chief executive Mustafa Suleyman's interview with Fortune: "people matter more than AI." Models built under the code must treat any violation of its provisions as a failure condition, meaning a system should abandon a task entirely rather than complete it in a way that breaches the rules. The code also establishes what Microsoft describes as a clear chain of command, intended to prevent an AI system from being used to enable large-scale harm such as cyberattacks or weapons development.
Specific prohibitions in the draft bar Microsoft's models from assisting with chemical, biological or nuclear weapons; offensive cyberattacks; mass-influence operations; child exploitation; nonconsensual deepfakes; or self-harm. Microsoft grouped the weapons-related restrictions under the acronym CBRNE, covering chemical, biological, radiological, nuclear and explosive threats, according to reporting from Decrypt and other outlets that reviewed the document.
A rejection of runaway AI
Suleyman used the announcement to draw a direct line under one of the AI industry's more speculative debates: whether advanced systems will eventually improve themselves beyond human control.
We shouldn't be trying to design models that can recursively self-improve beyond our control.
Mustafa Suleyman, Microsoft AI chief executive, to Fortune
Suleyman also pushed back on framing the debate around existential risk, arguing instead that alignment work should focus on ensuring "the purpose of technology is to serve humanity." Microsoft says a revised version of the code, informed by the public comment period, will follow before the end of the year and will continue to evolve as the company's MAI models grow more capable.
The timing is notable. Microsoft's draft arrives amid a broader industry reckoning over AI safety governance that has played out publicly in recent days, including debate over whether frontier labs should coordinate on safety standards and scrutiny of how quickly increasingly capable models are being deployed. Suleyman told Axios that "now's the time" for major AI labs to align on shared safety commitments, positioning Microsoft's code as both an internal governance document and an opening bid in that wider conversation.
Unlike a regulatory filing, the code of conduct is self-imposed and unenforceable by any outside body — its force depends entirely on Microsoft choosing to abide by it, and on public pressure generated by the comment process. Whether Microsoft's peers publish comparable commitments, and whether Microsoft's own MAI models can be shown to hold to the code as they scale, will be the tests that determine if Monday's draft amounts to more than a public-relations document.
