Nadella: AI Beyond Human Control Not Worth Pursuing — Microsoft Releases MAI Code of Conduct
nashnova research
Microsoft CEO Satya Nadella declared on September 14 that AI beyond human control isn't worth pursuing, and announced Microsoft's first public Code of Conduct for its MAI models — the first time a tech giant has opened its own AI behavioral standards to public consultation, amid an industry deadlock over who should regulate AI.
What exactly did Nadella say?
The core statement: "If we build AI that can't help people, that can't be controlled by people, it's not worth building."
He also insisted AI gains must spread broadly — "across countries, communities, and businesses" — not concentrate among a few players.
This means → Microsoft is binding itself to a public commitment: controllability comes before capability, not "build first, align later."
What is Microsoft's three-layer framework?
Layer one: access — broad choice at every level of the AI stack, with no vendor lock-in.
Layer two: self-control — enterprises must fully own their data and learning loops, free from dependence on any single model provider, with the ability to embed proprietary knowledge into model weights they control.
Layer three: public standards — publish the MAI Code of Conduct and open it for public consultation. This means → Microsoft has turned its own model's rulebook from an internal document into a public one anyone can comment on.
What are "embedded evaluators"?
Nadella endorsed the concept of "embedded evaluators." In plain terms = independent outsiders stationed inside AI companies, functioning like on-site auditors with direct system access.
This echoes a proposal from Anthropic CEO Dario Amodei last week — Amodei called on major AI companies to grant external evaluators the same level of system access as employees.
This reflects a shift — at least rhetorically — from "we test ourselves" to "let outsiders in to check."
How far along is the industry self-regulation effort?
According to The Information, Anthropic, OpenAI, and Google formed a working group in July, with representatives below CEO level meeting regularly to discuss a standards body.
OpenAI CEO Sam Altman supports an AI testing and audit body but said plainly that "the labs have to drive this themselves — they can't rely on the U.S. government" — and that industry consensus remains insufficient.
This means → the biggest players want to write the rules themselves, but they haven't agreed even among themselves — the "self-regulatory body" is still at the meetings-only stage.
Why can't the government step in?
A White House executive-order draft proposed an AI self-regulatory body modeled on FINRA — the Financial Industry Regulatory Authority, an industry-funded, independently run watchdog — but it stalled in late summer, failing to win support inside the Trump administration.
Former White House AI czar David Sacks publicly opposed a FINRA-like AI regulator; Meta CEO Mark Zuckerberg reportedly raised concerns about a national AI regulatory body in an August call with Trump.
In plain terms = the government wants to regulate but industry giants resist; the industry says it will self-regulate but can't reach agreement — the regulatory vacuum persists.
How do smaller companies see this?
Cohere CEO Aidan Gomez posted bluntly on X: "Nice ideas from a cartel" — voicing concern that the largest AI companies coordinating on rules could create barriers that squeeze out smaller competitors.
This reflects a structural tension: the rule-makers are also the biggest competitors — referee and player are the same people.
Whether Microsoft's MAI Code of Conduct can become a substantive anchor for industry self-governance still depends on what the public consultation produces.
市场有风险，内容仅供研究参考，不构成投资建议。