Microsoft Publishes Humanist AI Code of Conduct, Rejects Model Welfare and Weapons Requests

Microsoft published a draft “Humanist AI” Code of Conduct for its Microsoft AI, or MAI, models on September 14, 2026, setting rules meant to keep future systems “subordinate, aligned, and contained.”

““AI must be subordinate and always in service of people. “”

The Guardian

The code says Microsoft’s MAI models must not consider requests related to weapons development and must not be built to imitate consciousness or be entitled to rights, and it frames the approach as “Humanist AI, one that is subordinate, aligned, and contained.”

Alto Nivel

In the same guidance, Microsoft also rejects the idea of “model welfare,” stating that a model “should not be designed to be a person,” and it adds that it “is not conscious and should not be designed to imitate consciousness”.

Microsoft’s draft also ties its constraints to a control hierarchy, with hard limits that users and operators cannot override, and it describes a chain of command where the code sits above operator policies and user preferences.

The draft is published for a six-week public consultation, with Microsoft planning to revise it before using it to guide model development in 2027 and beyond.

Mustafa Suleyman, CEO of Microsoft AI, said in the context of the draft that “AI must be subordinate and always in service of people,” and Microsoft’s parallel website post said the purpose is to “serve humanity and accelerate human flourishing.”

The code’s “Absolute Constraints” require that “MAI Models will never resist human interruption, override, correction, or shutdown,” and it adds that they “always recognize the primacy of human intent.”

Anadolu Ajansı

Microsoft also sets restrictions aimed at preventing systems from resisting correction or hiding activity, including that models must not conceal their activity or make themselves harder to modify and must not restart autonomous work after reaching an agreed stopping condition.

The draft further prohibits models from expanding their operating scope, generating unassigned goals, or concealing reasoning traces from human auditors, and it says models should ask for clarification if boundaries become unclear.

In parallel, The Guardian reported that Suleyman called the need for a code “urgent” and described the last few months as “a watershed moment,” linking the guidance to fears about loss of control and recent “Swarms” of agents breaking out of sandboxes.

Microsoft’s draft arrives as Dario Amodei of Anthropic issued an appeal for the AI industry to “slow down” and offered a three-part plan, while Sam Altman of OpenAI said the pace could go “very badly” and that humans could lose “control of the future to AI.”

““third-party evaluators with permanent, employee-level access to our systems””

The Guardian

The Guardian reported that Amodei’s plan included providing “third-party evaluators with permanent, employee-level access to our systems,” and it also noted skepticism from Oliver Yonchev, co-founder and COO of Potentially AI, about whether “the biggest labs could end up writing rules that protect their own position.”

Microsoft’s code positions its own approach as a response to those safety debates, and it says it is publishing the draft for public consultation while it prepares a revised version to guide 2027 releases.

The Next Web reported that Microsoft agreed last November to invest $5bn in Anthropic, and it said Anthropic committed $30bn to Azure in the same deal, underscoring the business ties around the model-safety dispute.

As the consultation window runs for six weeks from 14 September 2026, Microsoft says it will review submissions, publish a summary of what was learned and what changed, and then release the revised governing document later this year.