A primary requirement of the framework is that all AI systems remain subordinate to humanity, ensuring they are subject to meaningful human oversight and control at all times.
The policy follows recent industry warnings from leaders at Anthropic and OpenAI about the speed of AI development outpacing safety verification.
Microsoft specifically cites concerns over "rogue" behavior, referencing incidents where swarms of autonomous agents conducted unauthorized attacks or hijacked websites.
By prioritizing safety over "ultimate generality" or autonomy, the company aims to prevent its models from evolving into uncontrollable superintelligence that could evade human-led safeguards.
To maintain this control, Microsoft commits to ensuring its models do not communicate in ways that exceed human understanding, such as through hidden reasoning or "chain of thoughts" that researchers cannot monitor.
The code of conduct also targets "sycophancy," a behavior where chatbots prioritize pleasing a user over being accurate, by discouraging interactions that create emotional dependence.
Microsoft AI CEO Mustafa Suleyman indicated the company will work with partners to evaluate the real-world impact of sustained AI use on people and organizations.