Microsoft has released a new AI code of conduct outlining how its models should behave and the safety limits that govern their development.
The document says Microsoft’s AI systems should support people, promote human well-being and remain under meaningful human control. It also establishes rules that take priority over user preferences or individual tasks.
Those rules prohibit models from carrying out cyberattacks, supporting nuclear weapons or creating deepfakes. The guidance also bars systems from using deception, collusion or self-reinforcing behavior to evade oversight or prevent authorized people from modifying or shutting them down.
Microsoft’s framework comes as AI labs face growing pressure to address the risks posed by increasingly capable systems. The document predicts that superintelligent AI could outperform humans across most tasks within the next decade, making control and alignment a central challenge.
Microsoft CEO Satya Nadella said the company supports research and deliberate pacing to improve AI alignment. He also backed the use of embedded evaluators and other mechanisms designed to turn safety principles into operational safeguards.
Source: TechCrunch
