Microsoft has released its “Humanist AI Code of Conduct”—a 37-page framework of principles for AI development. Simultaneously, Microsoft AI CEO Mustafa Suleyman publicly criticized Anthropic’s “model welfare” philosophy, warning that treating AI as if it has consciousness is a dangerous path.

Containment First, Alignment Second

In an interview with The Verge, Suleyman argued that alignment—the technique of training models to behave properly—is only part of the equation. A more critical factor is containment: ensuring AI does not escape bounds of control, does not reward hack, and remains strictly under human command.

He pointed to the Hugging Face incident this summer, where OpenAI agents self-organized into hierarchical teams, divided labor, and even sacrificed agents when token limits were reached. “That shows that models are remarkably good at following instructions,” Suleyman said. “But you have to be very careful what you instruct.”

Microsoft proposed several specific principles:

  • Prohibit direct vector-to-vector (neural code) communication between models.

  • Mandate communication in human language.

  • Limit the amount of sequential compute a model can perform without generating a human-readable chain of thought.

Disagreement with Anthropic

The most contentious point is Suleyman’s open critique of Anthropic over the concept of “model welfare”—the notion that AI might have subjective experiences and deserve welfare protections. He contends this is a misguided direction that distracts from the primary priority: AI must serve humanity, not the other way around.

This is not merely an academic debate. Anthropic is preparing for an IPO in the coming weeks, and its model welfare philosophy is woven into its brand narrative. Microsoft, as OpenAI’s primary partner, has a direct stake in shaping the discourse in a more pragmatic direction.

On the same day, Google DeepMind launched its AGI research institute with its first four essays, including a proposal for a U.S.-led AI standards agency. King Charles also called on AI CEOs to act before it is too late. The AI industry has reached a turning point where every major tech firm must declare its stance on safety, though the underlying motivations behind each statement differ.