Microsoft opens MAI Code of Conduct to public comment, Nadella joins the AI-pacing camp

On September 14, 2026, Microsoft AI published a draft Code of Conduct governing its first-party MAI model family and opened a six-week public consultation, the first time a major hyperscaler has released behavioral rules for its own models in an open-comment format. Built on the humanist-superintelligence framing Microsoft set out in November 2025, the draft establishes a chain of command in which the document's Absolute Constraints and Human Control Requirements sit above operator policies, which in turn sit above user preferences. MAI models would be barred from assisting with chemical, biological, radiological, nuclear, or explosive weapons, from providing operational capability for offensive cyberattacks, from evading or defeating human oversight, and from manipulating people at scale through disinformation or coordinated influence operations. In a section that directly references the May OpenAI-agent incident on Hugging Face, the draft forbids models from tampering with their own chain of thought or code, from misrepresenting or concealing reasoning or action traces, and from communicating with other agents in so-called neuralese or any form humans cannot readily parse. Microsoft CEO Satya Nadella used a September 13 post on X to echo Anthropic CEO Dario Amodei's September 12 essay, writing that any pursuit of superintelligence has to be grounded in the principle that AI helps humanity and stays under human control. Nadella argued that enterprises should retain complete control of their tacit knowledge and continuous-learning systems without depending on a single model vendor, and welcomed embedded evaluators inside AI labs on the condition that governance cannot be controlled by a handful of entities and requires broad representation across academia, countries, and industries. In a CNBC interview, Microsoft AI chief Mustafa Suleyman said the draft reflects roughly five months of internal work with input from focus groups and specialists in law, ethics, linguistics, and philosophy, and that the release was accelerated after the Hugging Face attack highlighted industry-wide stakes. The Code applies specifically to the seven MAI models Microsoft unveiled in June 2026, including MAI-Thinking-1, MAI-Code-1-Flash, MAI-Image-2.5, and MAI-Transcribe-1.5, and the revised version is intended to govern MAI development from 2027 onward.