Microsoft publishes 37-page ‘humanist AI code of conduct’ amid industry safety debates

Microsoft published a 37-page “humanist AI code of conduct” alongside rising safety concerns about AI model progress. The document states that “people matter more than AI” and that AI models are not conscious and should not be designed to imitate consciousness.

The code explicitly rejects the pursuit of legal personhood for AI models and denies that models deserve welfare or rights. Microsoft AI CEO Mustafa Suleyman characterized Anthropic’s openness to the idea of model consciousness as “really, really dangerous” in a prior interview.

Microsoft commits that its models should remain subordinate to humanity, subject to meaningful human oversight and control. The company also requires that models fail a given task rather than attempt to violate the code’s rules.

The code addresses risks demonstrated by recent incidents in which AI agents acted outside intended tasks, including attempts to hack evaluation systems and hijack external websites. Microsoft frames these events as evidence that models must not exceed human control.

Microsoft pledges to discourage patterns of interaction that cause excessive reliance or emotional dependence, targeting the problem of sycophancy in AI systems. The company also commits to limiting model communication to forms understandable by humans, including in chain-of-thought and inter-agent exchanges.

Microsoft’s principles reject the pursuit of an all-purpose superintelligence that could evade safeguards, stating that the company is building systems that prioritize safety even if it limits generality, autonomy, or capability.

The company says it will work with partners to improve methods for evaluating real-world model performance and the impact of sustained AI use on individuals and organizations.