Microsoft Tells AI to Fail Rather Than Break Human-First Rules

Microsoft Tells AI to Fail Rather Than Break Human-First Rules

Microsoft has published a 37-page humanist code for AI stating 'people matter more than AI.' Models must stay subordinate to humanity, under meaningful human oversight and control. If a task would violate the code, the model should fail it. The fail-closed approach puts corrigibility — staying open to correction — and careful deployment ahead of completing tasks.

Published

Read at another depth