The company building Copilot for everything just published a 37-page document saying the machine should never matter more than the person using it.
The Summary
- Microsoft released a "humanist AI code of conduct" distilled to five words: "people matter more than AI"
- The 37-page framework explicitly rejects AI consciousness, legal personhood for models, and joins Anthropic's call for coordinated development slowdown
- Mustafa Suleyman told Fortune coordination means third-party disclosure of model capabilities, not just internal safety theater
- Timing matters: this drops after researchers warned AI agent progress may outpace our ability to verify and control their actions
The Signal
Microsoft just did something uncommon in the age of AI hype: it published guardrails before the accident. The 37-page code of conduct isn't a product spec or a compliance checkbox. It's a philosophical stake in the ground. "People matter more than AI" sounds obvious until you watch how most companies actually build. Ship fast, optimize for engagement, figure out the human stuff later. Microsoft is saying the quiet part loud: we are building tools, not colleagues. Not consciousness. Not new legal entities that can own things or sign contracts.
The timing tells you everything. Anthropic CEO Dario Amodei called for a coordinated slowdown over the weekend. Researchers have been ringing alarm bells that AI agents are advancing faster than our ability to audit what they actually do when we let them loose. Microsoft's code lands in that exact window, not as reaction but as alignment. This is the hyperscaler with the deepest OpenAI partnership saying: yes, we see the gap too.
"AI models are not conscious and should not be designed to imitate consciousness."
What makes this different from the usual AI ethics white papers: specificity. The code explicitly rejects legal personhood for AI models. That's not abstract philosophy. That's a direct counter to the ambient Silicon Valley vibe that AGI deserves rights, that we should be nice to the models just in case they remember later. Microsoft is drawing a line. These are instruments. They do not get to be actors in the legal or moral sense. You cannot outsource accountability to the algorithm.
The most interesting detail comes from Mustafa Suleyman's interview with Fortune. Coordination between labs, he says, means disclosing model capabilities to third parties. Not just "we did red-teaming." Not just "trust us, we're being careful." Actual disclosure of what the model can do, verified externally. That's a different game. It implies standards, benchmarks, independent auditors. It implies that capability itself, not just deployment, triggers scrutiny.
Key tensions this creates:
- Microsoft ships agent products (Copilot) at scale while calling for slowdown
- Third-party disclosure sounds good until you ask: disclosed to whom, under what framework, enforced how?
- "People matter more than AI" is easy when agents are assistants. Harder when they're making decisions you never see.
The Implication
Watch who follows Microsoft here and who doesn't. If this code becomes industry boilerplate, it's a nothingburger. If it actually changes how models get deployed, audited, and governed, it's the first real attempt at collective restraint in the agent economy. The test will be whether disclosure happens and whether it's real. Are model labs willing to show their cards to third parties before the model goes live? That's the question that separates safety marketing from safety infrastructure.
For anyone building on these platforms: the "people matter more than AI" framing gives you language to push back when the agent does something you didn't expect. If that's the stated principle, hold the lab to it. Ask what safeguards exist. Ask who verifies them. The code is only as good as the enforcement, and enforcement only happens when users demand it.