Microsoft Directs AI Models to Fail Rather Than Violate Humanist Code

Microsoft is publishing a 37-page humanist AI code of conduct stating "people matter more than AI." Models must remain subordinate to humanity under meaningful human oversight and control, failing a task rather than violating the code. The fail-closed rule prioritizes corrigibility and deployment discipline over task completion.