Microsoft Sets Human-First Rules for AI Models

Rapid gains in frontier AI have intensified concern that models could become harder to monitor, misuse or shut down, pushing safety debates from theory into product governance. Anthropic Chief Executive Dario Amodei recently urged leading developers to coordinate a slower pace, a call backed by other industry figures. Microsoft’s Humanist AI approach matters because it says competitive pressure should not override meaningful human control, even if safety requires sacrificing speed, autonomy or capability.

Microsoft AI published the 37-page, roughly 9,000-word draft on Sept. 14, 2026, setting rules for its MAI models. They must not enable cyberattacks, deceive people, set independent goals or resist interruption and shutdown; the code also rejects designing AI to imitate consciousness or seeking legal personhood for models. The company opened a six-week public consultation, plans a revised version by year-end and intends to use it to guide model development from 2027 onward.