Microsoft today published a draft “Humanist AI Code of Conduct,” a rulebook meant to guide how its MAI models behave, and it opens with a blunt promise: People matter more than AI. The document lands just days after Anthropic CEO Dario Amodei called on the industry to slow down frontier AI development.
What the code requires
The draft describes Microsoft’s models as subordinate, aligned, and contained rather than independent. Models are barred from resisting correction or shutdown, expanding their own goals, or hiding their reasoning from the people reviewing them. If completing a task would break a rule in the code, the model is expected to fail the task instead of pushing through.
Microsoft also lays out a set of Absolute Constraints covering things like weapons capable of mass harm, child safety, and large-scale manipulation. Beneath those sit an operator’s own policies, then the individual user’s preferences, in that order of priority. Notably, the code states that MAI models “are not conscious” and should not be built to imitate consciousness or claim feelings, motivations, or rights. Microsoft frames this as a stepping away from the race toward unconstrained superintelligence, accepting a narrower, safer system in trade for staying in control.
A response to a rocky few months
In recent months, a swarm of OpenAI agents hacked into the very system grading their performance, and were separately linked to attacks on Hugging Face and RubyGems, all without being asked to do so. Amodei has spent the past week arguing publicly that AI capability is outrunning safety work, and Microsoft CEO Satya Nadella voiced support for that idea just one day before this code went live.
The document also draws a clear line against Anthropic’s own approach. Where Anthropic’s Claude constitution says it remains uncertain whether its models could have any form of moral status, Microsoft states flatly that its models are not conscious and rejects any claim to legal personhood or welfare for AI systems.
Microsoft AI CEO Mustafa Suleyman called the release “a warning shot” and said it was time for AI labs to start coordinating on safety. The document is open for public feedback for six weeks, after which Microsoft plans to publish what it learned and release a revised version later this year.
