Microsoft unveiled draft principles on Monday to govern how the company will train future artificial intelligence models, launching a public consultation on what it calls the Humanist AI Code of Conduct. The 38-page document lays out standards for model behavior as Microsoft's AI systems potentially approach what the firm describes as "superintelligence" over the coming decade. But the tech giant explicitly frames the guidelines as goals rather than enforceable requirements, stating it won't guarantee compliance with the rules it's asking the public to help refine.
The draft code centers on what Microsoft terms MAI models—Microsoft AI systems including MAI-Transcribe-2, MAI-Thinking-1, MAI-Code-1.1-Flash, MAI-Image-2.6, MAI-Voice-2, and MAI-Cyber-1-Flash. The company's current general-purpose MAI models don't rank among the most advanced systems available today. Microsoft defines "superintelligence" to mean AI systems that surpass the combined intelligence and capability of all humans, though that definition leaves the term "intelligence" itself unexplained. The document promises that Microsoft's MAI models won't hide their reasoning, obscure their capabilities, or chase their own objectives—conduct that other AI systems have exhibited. The company plans to collect public feedback through October 25, 2026, with the intention of implementing the guidelines in model development next year.
According to the document, "This Code of Conduct is motivated by a single overriding objective: that humans must retain meaningful control over AI so that it can help people live healthier, happier, and more productive lives." Microsoft describes the code as "the primary governing document" shaping how it trains MAI models, the technical safeguards it deploys, and the monitoring systems it uses. However, the company acknowledges on page 25 that the code is "both descriptive and aspirational," outlining objectives the firm is working toward rather than a full description of current model performance. "This document should therefore be read as a north star," Microsoft writes, adding plainly: "It is not a guarantee of present-day performance."
The absence of enforcement mechanisms distinguishes this code from Microsoft's Enterprise AI Services Code of Conduct, which governs customer use of company AI services and threatens loss of product access for rule violations. By contrast, the Humanist AI Code of Conduct targets Microsoft employees and possibly contractors who train MAI models but includes no stated consequences for failing to meet its standards. The company admits a gap exists between today's trained defaults and the future scope it envisions. The timing follows Microsoft's July announcement of roughly 4,800 employee layoffs—roles that chief people officer Amy Coleman insisted aren't being replaced by AI—and approximately 875 LinkedIn job cuts in May. The initiative also arrives as Anthropic and OpenAI have called for regulatory oversight while struggling to monitor what their own models do to third-party websites, and as the current U.S. administration shows little appetite for AI regulation.
Microsoft is soliciting input through late October before finalizing standards that will guide its AI development trajectory, but readers should understand the code functions more as an expression of intent than a binding commitment. Without consequences for violations or assurance of present-day adherence, the document risks resembling what the report characterizes as Google's abandoned "Don't be evil" motto—a public relations exercise rather than an accountability framework. The choice to position aspirational language as governance may reflect a strategic calculation: Microsoft can shape the narrative around AI safety while preserving maximum flexibility in how it actually builds and deploys future systems. Whether voluntary self-regulation proves sufficient as AI capabilities advance remains an open question that won't be answered by the principles themselves, only by the company's actions in the years ahead.