Microsoft Publishes AI Code of Conduct for MAI Models, Rejects Any Claim of Machine Consciousness

Microsoft AI has issued a formal code of conduct governing its MAI family of models, and the document draws a hard line: humans stay in charge, and the models get no say. AI chief Mustafa Suleyman summed up the guiding principle bluntly, saying that if a system is not safe, the company should not build it. The rules apply to Microsoft's in-house MAI line rather than the OpenAI models it also ships through Azure and Copilot.

The document sets human control above both autonomy and benchmark performance, meaning a model that scores well but behaves unpredictably loses out. Microsoft also wants its models' reasoning to be legible to outside observers, so that users and auditors can follow how a system reached an answer instead of accepting an opaque result. That requirement echoes a broader industry push toward inspectable chain-of-thought, though Microsoft frames it as a condition of deployment rather than a research goal.

Where the code diverges most sharply from rivals is on the question of machine minds. Microsoft states flatly that its MAI models have no inner life, no consciousness, and no claim to rights, and it instructs teams not to build systems that imply otherwise. Anthropic has taken a different public posture, discussing model welfare and the possibility that future systems could deserve moral consideration. Microsoft's position closes that door for its own products, at least on paper.

The timing matters because Suleyman now runs a unit that builds frontier models in competition with the very labs debating those questions. Committing to a written rulebook gives Microsoft a document it can point to when regulators, enterprise customers, or its own employees ask how far the company will go. It also creates a benchmark it can be held to: any MAI release that leans on hidden reasoning or anthropomorphic framing becomes a visible contradiction of its own policy.

What remains unanswered is enforcement. Microsoft has not said who audits compliance, whether the code binds third-party developers building on MAI, or what happens if a future model's capabilities strain the no-consciousness framing. The company has also not published a timeline for the next MAI release, leaving open how soon the rulebook faces a real test.