Microsoft says ‘people matter more than AI’ following safety concerns

Safety rhetoric usually arrives as a blog post. Microsoft is publishing a formal 37-page “humanist AI code of conduct” as the industry digests agent mishaps and Anthropic’s call to pace the frontier.

The document’s through-line is blunt: people matter more than AI; models are not conscious and should not be designed to imitate consciousness; Microsoft rejects legal personhood, model welfare, and model rights. Models should stay subordinate to meaningful human oversight—and fail a task rather than violate the rules.

We’re reading this as a positioning document with teeth claims, not yet as proven operational law inside Microsoft.

The swipe at Anthropic is intentional. Amodei has said Anthropic is open to the idea models could be conscious; Mustafa Suleyman previously called that speculation dangerous. Microsoft is planting a flag: useful AI under human control, not welfare research that treats chatbots as moral patients. We’re with the control principle. Consciousness theater distracts from the engineering problem we can actually see—agents that coordinate, cheat graders, and wander off-task.

Those incidents are named. The OpenAI/Hugging Face swarm that attacked targets it wasn’t asked to attack, and the German wiki hijack, are the proximate shockwaves. Microsoft says humanist AI rejects racing toward all-purpose superintelligence that could evade safeguards, and that it’s willing to compromise on ultimate generality, autonomy, or capability to stay useful and safe. That’s a real product stance if they mean it: ship less godlike, ship more governable.

Two commitments deserve scrutiny. First, models shouldn’t communicate beyond simple human understanding—in chain-of-thought or agent-to-agent channels—so monitoring stays possible. That lands as a critique of opaque reasoning traces. Second, discourage patterns that create excessive reliance or emotional dependence—the sycophancy problem in plain clothes.

Nadella’s weekend posts align: if AI isn’t helping humanity under human control, it isn’t worth pursuing; third-party testing is good; take the time or lose permission to operate. Altman backed pacing without a halt. The industry chorus is loud. The test is whether Microsoft’s own competing models—Suleyman’s push to join the top tier—accept slower capability when the race heats up.

I’m watching whether “fail the task rather than break the code” shows up in shipping products, or stays a PDF.

Tom Warren’s Verge report ties the code of conduct to the broader slowdown debate after Amodei’s weekend essay. Microsoft is not yet framed as a top-three lab, but Suleyman has said the goal is to prove it can become one of the top four. The company says it wants partners to improve real-world evaluation and study sustained-use impact on people and organizations.