Something significant happened in the AI industry this weekend, and if you are running a business on Microsoft tools, it is worth paying attention.

On September 13, Satya Nadella announced that Microsoft would publish a Code of Conduct governing its first-party MAI models, opening it for public consultation the following day. The announcement came hours after every major AI lab CEO — Dario Amodei at Anthropic, Sam Altman at OpenAI, Demis Hassabis at Google DeepMind, and Elon Musk at xAI — publicly endorsed a proposal to deliberately slow the pace of AI capability development.

This is not just an internal policy document. It is the first time Microsoft has codified how its own AI models are supposed to behave, and it is being published for the world to comment on.

What Nadella Actually Said

The announcement came via a post on X (formerly Twitter), where Nadella laid out Microsoft’s three-part approach to AI development:

• Broad access and choice at every layer of the AI stack
• Enterprise control of learning loops and models
• A Code of Conduct underlying Microsoft’s first-party MAI models

His framing was direct: “Any pursuit of superintelligence has to be grounded in the core principle that if the AI we build is not helping humanity and under human control, it’s not worth pursuing.”

Microsoft is endorsing what AI labs are calling “deliberate pacing” — slowing the rate of capability improvements to give safety research time to keep up — along with “embedded evaluators,” independent oversight figures who would sit inside AI companies with access to models, tools, and internal researchers.

The Bigger Picture Behind the Announcement

This did not happen in isolation. On September 12, Anthropic CEO Dario Amodei published a roughly 3,800-word essay titled “We Must Pace the Frontier,” arguing that AI advances are moving faster than researchers’ ability to understand and control them. Within hours, Sam Altman (OpenAI), Demis Hassabis (Google DeepMind), and Elon Musk (xAI) all publicly endorsed the idea.

Separate reporting from The Information revealed that Anthropic, OpenAI, and Google DeepMind have been quietly holding working-group meetings since July to design an industry-led standards body — essentially a governing body for AI safety, modelled in part on FINRA’s role in financial markets.

The consensus that formed this week across competing AI labs is unusual. These companies fight aggressively for talent, customers, and benchmark rankings. Agreement on anything is rare.

It has not been without controversy. David Sacks, a former White House AI adviser, publicly warned the labs that coordinating on standards could cross into antitrust territory, rejecting any government waiver that might allow such coordination and questioning whether safety is the real motivation.

What This Means for Businesses Using AI

For business leaders, there are three practical implications worth thinking through.

You now have something to hold Microsoft accountable to. A published Code of Conduct for AI model behavior is a reference document. When a MAI model inside Copilot or Azure does something unexpected, you will have a baseline to reference. That is a meaningful shift from the current situation where model behavior is governed by terms of service buried in licensing agreements.

“Enterprise control of learning loops and models” is the key phrase. Nadella specifically included this in his three-part framework. This signals that Microsoft is moving toward giving enterprise customers more control over how AI models learn from their data and adapt to their environment. If you have been worried about what happens to your company data when it flows through Copilot, this is the direction the conversation is heading.

The deliberate pacing trend means model capability improvements may slow. If the major labs follow through on their pacing commitments, the sprint of capability improvements that has defined the past two years could slow down. For businesses trying to build on a moving platform, that could actually be welcome news. A more stable capability baseline makes enterprise AI deployments easier to plan and maintain.

The flip side is that anyone hoping for a GPT-7 or Fable 6 to solve their current AI limitations by next quarter should factor in that timeline may be longer than expected.

What Stays the Same

The pacing debate is mostly about frontier model development — the cutting edge of what AI can do. It is separate from the deployment and application layer where most business value is created.

The agents, automations, voice AI systems, and analytics tools that are already deployed are not going anywhere. If anything, a period of slower capability growth combined with stronger governance frameworks creates better conditions for enterprises to actually operationalise the AI investments they have already made, rather than constantly chasing the next model release.

The real risk in 2026 is not getting left behind by the next model. It is having deployed AI systems with no governance, no oversight, and no accountability. Microsoft’s Code of Conduct, whatever its final form, is a step toward the kind of infrastructure that makes enterprise AI deployable at scale.

What to Watch Next

The public consultation on Microsoft’s MAI Code of Conduct is open as of today. It is worth reading when published. The industry-led standards body discussions between Anthropic, OpenAI, and Google are ongoing. And the White House antitrust concern is a genuine complication — if government blocks the labs from coordinating on safety, the alternative is fragmented, competing standards that enterprise buyers will have to navigate independently.

For businesses building on AI now, the governance conversation is no longer abstract. It is happening at the CEO level across every major AI company in the world.

---

Enterprise DNA helps businesses build the data and AI foundations they need to move faster and operate smarter. Explore Omni services or build your team’s AI skills.