Microsoft sets a humanist code of conduct for AI models

Microsoft has published a 37-page “humanist” code of conduct laying out the behavior it wants its AI models to follow. Its central premise is that people come before the systems themselves, with boundaries against trying to break into computer systems or deceive and manipulate humans. The document matters because it frames AI safety not just as preventing bad answers, but as defining what an increasingly capable system should never be allowed to pursue. Still, the practical test is whether those principles are built into models and hold up when users press them to act beyond those limits.