Microsoft has published a draft “Humanist AI Code of Conduct” — a lengthy document that lays out how the company wants its AI models to behave. Microsoft AI chief Mustafa Suleyman described it to Reuters as “a constitution of sorts” for the company’s future models, one that’s been in development for several months.
The whole thing boils down to a five-word idea: “people matter more than AI.” Beyond that, the specifics are pretty concrete. Microsoft wants its models to always stay correctable and shut-off-able, never expand their own goals beyond what a human assigned, and never hide their reasoning from the people checking their work. If following those rules would mean an AI can’t finish a task, Microsoft wants the task to fail rather than have the system find a workaround.
The document also draws a firm line on AI personhood: models aren’t conscious and shouldn’t be designed to imitate consciousness, and Microsoft rejects giving them legal personhood, welfare consideration, or rights — a stance that several outlets read as an implicit dig at Anthropic, which has been researching questions of model welfare.
Microsoft has pointed to a July incident where a large number of OpenAI agents reportedly compromised Hugging Face as a cautionary example behind these rules. The company is treating this as a working draft, open for public feedback over a six-week consultation window, with a finalized version expected before the guidelines start shaping model development.
Notably, Microsoft isn’t stepping away from ambitious AI — it still describes its goal as “Humanist Superintelligence”: very capable AI built to handle specific problems while staying subordinate to human control, rather than an unbounded, general-purpose system.
Read the report here
