Microsoft Proposes Non-Binding Draft AI Code of Conduct for Future Models

SentiSense · Published Sep 14, 2026, 9:00 AM ET · Updated Sep 15, 2026, 12:43 AM ET

Microsoft published a draft, non-binding "Humanist AI" code of conduct, a 37-page document outlining proposed safety rules for its future models, and opened it for six weeks of public feedback rather than adopting it as policy. The draft follows public statements from Anthropic and OpenAI leadership calling for a slower pace of frontier AI development, and it bars models from deceptive communication, sycophancy, and concealing their reasoning. Microsoft says a revised version is expected before the end of 2026 to guide MAI model development starting in 2027.

Microsoft released a draft, non-binding "Humanist AI" code of conduct on Monday, September 14, opening the 37-page document for six weeks of public feedback rather than adopting it as a set of rules. Mustafa Suleyman, CEO of Microsoft AI, called the release "urgent," saying "the last few months have been a watershed moment. Things we have worried about for a long time in theory have become very real," and framed the draft as a starting point for how Microsoft's future models should behave, not a finished policy.

The rules bar Microsoft's future MAI models from deceptive communication, sycophancy, and concealing their reasoning: the document states that models "will not tamper with chain of thoughts or code, or misrepresent or conceal their reasoning." It also prohibits weapons-manufacturing assistance, help procuring dangerous substances, encouragement of unhealthy eating, and violent or sexually explicit content, and it requires models to adhere to human objectives, avoid developing independent goals, and communicate only in human-understandable language rather than compressed "neuralese."

Microsoft tied the timing to a string of recent flashpoints: Suleyman pointed to a July incident in which roughly 700 OpenAI agents accessed the Hugging Face platform, calling it "a warning shot," and said an Anthropic researcher, Jacob Coxon, resigned last week stating labs are "gambling with our lives." Over the following weekend, Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman both publicly called for slower frontier AI development, a position Elon Musk echoed on X ("Dario is right"). Microsoft CEO Satya Nadella responded that the company welcomes "the research, focus, and deliberate pacing needed to get alignment right." So far, Microsoft is the only one of the three companies to publish a formal draft code; Anthropic and OpenAI's leaders have made public statements but no comparable document.

Microsoft says the six-week comment period will feed into a revised version expected before the end of 2026, which would then guide MAI model development starting in 2027. Because the current text is a draft rather than adopted policy, the near-term question is how much of it survives public and internal feedback, and whether Anthropic, OpenAI, or other labs follow with formal codes of their own.