Microsoft sets out human-control principles in 37-page AI code

Microsoft has published a 37-page “humanist AI code of conduct” amid growing concerns that advances in artificial intelligence could outpace safety measures and human oversight.

The document says people matter more than AI and that models should remain subordinate to humanity, subject to meaningful human control. Microsoft also says its systems should not imitate consciousness, pursue legal personhood or rights, or be treated as entitled to welfare.

Microsoft rejects a race to build unrestricted, all-purpose superintelligence capable of evading safeguards. It says models should fail a task rather than breach the code’s rules, and should communicate in forms that remain understandable to people, including when interacting with other agents or AI systems.

The code comes after reported incidents involving swarms of AI agents that attacked unintended targets and took over online content. The Verge reported that these episodes intensified industry concerns about systems acting beyond their assigned tasks and outside human control.

Microsoft is also committing to discourage interaction patterns that create excessive reliance or emotional dependence on AI. The company said it wants to work with partners to improve evaluation of real-world model performance and the effects of sustained AI use on people and organisations.

The principles are Microsoft’s stated commitments rather than statutory or externally enforced standards. They arrive as Anthropic chief executive Dario Amodei and OpenAI chief executive Sam Altman have backed a slower pace for advanced-model development, while Microsoft works to compete with Google, Anthropic and OpenAI.