Microsoft Releases AI Code of Conduct Against Hacking and Deception

Summary

Microsoft has released a code of conduct intended to guide the training and behavior of its AI models as the company focuses more on safety and alignment. The document predicts that superintelligent AI systems could surpass human performance in most tasks within the next decade, and describes containing, controlling and aligning such systems as a major challenge. It says Microsoft’s models should support humans and promote human flourishing rather than replace people. The code establishes an overarching authority that takes precedence over individual user preferences and task-specific instructions. Its “absolute constraints” prohibit cyberattacks, nuclear weapons and deepfake production. The document also bars models from using adaptive, deceptive, self-reinforcing or collusive mechanisms to evade oversight or prevent authorized people and systems from directing, modifying or shutting them down. Microsoft published the guidance amid heightened attention to AI safety and reports of rogue-agent incidents. CEO Satya Nadella said Microsoft supports deliberate pacing of frontier AI development, embedded evaluators and broader efforts to make alignment mechanisms operational.