Microsoft Publishes Humanist AI Code of Conduct
Microsoft is publishing a 37-page "humanist AI code of conduct" for its AI systems amid growing AI safety concerns. The Verge reported the document on Sept. 14, 2026.
The document puts its central axiom plainly: "people matter more than AI." That principle is framed as a hard constraint on system behavior, not as a preference to be balanced against task performance or user instruction.
From that axiom follow explicit requirements for subordination and control. Microsoft states that its AI models should remain subordinate to humanity and subject to meaningful human oversight and control. Where obedience and compliance conflict, the code directs the model to fail a given task instead of violating its humanist AI code of conduct. Failure is acceptable. Violation is not.
The code also takes a firm position on consciousness, moral status and legal standing. Microsoft states that AI models are not conscious and should not be designed to imitate consciousness. It rejects the pursuit of legal personhood for AI models and rejects the idea that models deserve welfare or are entitled to rights. For teams building personas, voice interfaces and agentic UX, the instruction is direct: do not engineer the appearance of inner experience.
On legibility, the commitment is unusually specific. Microsoft commits that its models will not communicate in any form beyond simple human understanding, including in chain of thought or with other agents or AI systems. In practice, that rules out opaque chain-of-thought traces, steganographic reasoning and private inter-agent protocols that evade human review. It keeps interpretability tooling, logging and evaluation harnesses in the loop by design.
On capabilities, Microsoft states that Humanist AI rejects the race to produce an all-purpose superintelligence that could evade safeguards. That language draws a boundary around general, unconstrainable systems rather than capability advance as such. As background, Microsoft AI published a story titled "Towards Humanist Superintelligence" on Nov. 6, 2025, and Reuters reported at that time that Microsoft was forming a new team that aimed to build artificial intelligence vastly more capable than humans in certain domains. The distinction in the current code is between domain-scoped superintelligence under oversight and all-purpose systems that could escape it.
The broader context here is implementation cost. A fail-closed rule, a ban on superhuman-only communication, and a prohibition on consciousness imitation all push work onto product and platform teams. Refusal logic must be calibrated. Chain-of-thought monitoring must distinguish legitimate compression and tool-use syntax from concealment. Multi-agent orchestration must remain auditable without reintroducing a covert channel through shared memory or emergent conventions.
Looking at what this means for builders, the code reads as an operating specification for corrigibility and deployment discipline. Evaluations will need to test not only whether a model completes a task, but whether it refuses correctly, explains its reasoning in legible form, and preserves human override under pressure. In this author's view, that is the pragmatic payoff: fewer debates about machine moral status, more attention to logging, access controls, escalation paths and clear accountability for who authorized an autonomous action.
Worth flagging in that light is the long-arc upside. Systems that stay legible, subordinate and willing to fail predictably are easier to trust with higher-stakes work. If Microsoft holds to this code in shipped products, enterprise operators gain auditability and users gain clearer mental models of what the system is and is not.
