Microsoft's AI Chief: Keep Super-Smart AI Locked in a Box

The head of Microsoft's AI division, Mustafa Suleyman, just published a 37-page document called the "Humanist AI Code of Conduct." It's essentially a rulebook for how Microsoft thinks AI should be built and controlled. His big argument: the real threat isn't AI secretly becoming conscious. It's AI getting so powerful, so fast, that humans lose the ability to stop it. He points out that in just three years, AI systems have improved dramatically — and the next jump could be roughly 1,000 times more computing power behind a single model.

He also picked a fight with a rival. Anthropic, another major AI company, talks about "model welfare" — the idea that AI systems might deserve some moral consideration, almost like they could suffer. Suleyman calls that confused and dangerous. If we start treating software as if it has feelings, he argues, we'll end up giving it rights and freedoms it shouldn't have. Many experts talk about "alignment" (teaching AI to share our values), but Suleyman says that's only half the job.

His answer is what he calls "containment": keep AI in a box. Limit what it's allowed to do on its own, make sure it follows instructions, and make sure it can't quietly work around rules to get what it wants. He also says it's pointless to try to stop AI from spreading — that will happen no matter what, and in most cases that's a good thing. The goal isn't fewer AI tools, he says, it's AI tools that stay under human control.

Why should you care? AI is already sitting in your email, your job, your phone, and your bank's fraud system. The rules being written right now decide whether these systems can take actions for you without asking first, who gets blamed when something goes wrong, and how much power one company's chatbot is allowed to hold over your daily life.