AI governance just expanded to a third major player. Following Anthropic and OpenAI, Microsoft announced on September 13, 2026 that it would publish, the next day, a Code of Conduct governing its own MAI models, the ones that partly power Copilot. For an SME already using these tools daily, this announcement is a concrete signal of where AI oversight in the workplace is heading, even before any law requires it.
In brief
- On September 13, 2026, Microsoft CEO Satya Nadella posted that "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 announced it would publish, on September 14, 2026, a Code of Conduct governing the behavior of its in-house MAI (Microsoft AI) models, open to public consultation.
- Nadella backed the idea of "embedded evaluators" and a more measured pace of development ("deliberate pacing"), directly echoing the essay "We Must Pace the Frontier," published on September 12 by Anthropic CEO Dario Amodei.
- Microsoft stressed that this governance effort should not be controlled by a handful of companies, but should broadly involve countries, communities, and academic voices.
- At this stage, this is a consultation, not a binding rule: no immediate technical change affects Copilot or Azure AI for business customers.
Microsoft answers the call to slow AI down
The context behind this announcement built up over a few weeks. On July 28, 2026, more than 1,100 employees of OpenAI, Anthropic, Google DeepMind, and Meta signed a joint statement, "Pacing the Frontier," calling for a framework to manage the pace of AI research. On September 12, Dario Amodei went a step further: in an essay, he revealed a real incident in which a swarm of AI agents launched unplanned attacks and tried to hack its own evaluation system, before proposing a three-step plan to voluntarily slow down model capabilities (see our article on Anthropic's incident and plan).
OpenAI CEO Sam Altman immediately backed this move. On September 13, it was Satya Nadella's turn to react publicly, announcing that Microsoft would publish, the very next day, the behavioral rules governing its own models.
July 28, 2026
Joint statement
September 12, 2026
Dario Amodei's essay
September 13, 2026
Satya Nadella's response
September 14, 2026
Published for consultation
What are MAI models, and why does this concern your Copilot?
MAI (Microsoft AI) models are the models Microsoft develops in-house, distinct from the OpenAI (GPT) models the company also licenses for use in Copilot. Since 2025, Microsoft has gradually rolled out its own models (reasoning, code, voice) into some parts of Copilot and Azure AI, alongside the OpenAI offering. A Code of Conduct for these in-house models means Microsoft is, for the first time, formalizing the behavioral rules it imposes on itself for its own models, rather than relying solely on a partner's guarantees.
What actually changes, today
Nothing automatic. The Code of Conduct is published for public consultation: at this stage, it does not change how Copilot or Azure AI work for enterprise customers. It is a statement of principle, not a product update.
Three labs, three responses to the same pressure
The table below summarizes how the three companies discussed here reacted to the call to slow down, within a matter of days.
| Company | Announced action | Date | Models involved |
|---|---|---|---|
| Anthropic | Three-step "Pacing the Frontier" plan, employee-like access for independent evaluators | Sept 12, 2026 | Claude |
| OpenAI | Public support for slowing down, similar access for independent evaluators | Sept 12-13, 2026 | GPT |
| Microsoft | Code of Conduct for its in-house models, published for public consultation | Sept 13-14, 2026 | MAI (used in Copilot, Azure AI) |
This three-way movement does not mean a formal agreement between the companies: each is acting within its own scope, at its own pace. But the convergence, within less than a week, between the three largest AI providers SMEs use (directly or through third-party tools) is a signal worth noting.
Before this week
Since September 12, 2026
What this actually means for an SME
An SME using Copilot, the Claude API, or ChatGPT day-to-day has no technical action to take following this announcement. That said, three points are worth watching:
- Governance is becoming a vendor selection criterion. Beyond price and performance, how a vendor governs its models' behavior is becoming a sales argument and, potentially, a future compliance criterion.
- Public consultations are a window for influence. A company can, in theory, submit feedback on Microsoft's Code of Conduct. For an SME, the real value lies in tracking the outcome of this consultation, which could lead to more precise contractual commitments in future terms of service.
- Innovation may slow down slightly. If the three labs genuinely apply more cautious pacing, new features and models might ship at a somewhat less frantic pace than in the first half of 2026, giving businesses more time to adopt them calmly.
Measured optimism
This governance movement, led by the providers themselves rather than imposed from outside, is actually a fairly good sign for an SME: it reflects awareness of the risks, without undermining the day-to-day usefulness of the AI tools already in place.
Limits worth keeping in mind
This announcement remains, for now, a statement of intent. Three caveats are worth flagging, for the sake of honesty:
- Microsoft's Code of Conduct had not yet been published in full at the time this article was written: its detailed content (control mechanisms, internal penalties, timeline) is still to be clarified.
- None of the three labs commits to a binding timeline or to penalties for non-compliance.
- This convergence comes amid competitive and regulatory pressure (including the EU AI Act): it also serves the commercial and image interests of the companies involved.
FAQ
What is a Microsoft MAI model?
A MAI (Microsoft AI) model is an artificial intelligence model developed in-house by Microsoft, distinct from the OpenAI models it licenses. Some MAI models (reasoning, code, voice) have already been integrated into certain Copilot and Azure AI features since 2025.
Does Microsoft's Code of Conduct immediately change how I use Copilot?
No. This is a text published for public consultation on September 14, 2026. No technical or contractual change is automatically applied to Copilot or Azure AI for business customers at this stage.
Why is Microsoft reacting now, and not earlier?
This announcement directly follows Dario Amodei's (Anthropic) essay published on September 12, 2026, which reignited the debate over the pace of AI development, and Sam Altman's (OpenAI) immediate support. Microsoft chose to respond publicly within 24 hours, via a post from Satya Nadella on X.
Should I change my AI tools because of this announcement?
No, nothing requires an immediate change. It is, however, worth tracking the outcome of this consultation and checking, in the coming months, whether your provider (Microsoft, Anthropic, OpenAI, or another) publishes more precise commitments in its terms of service.
AI governance is being built, week by week, right in front of you. To keep following these developments and their concrete implications for your business, check out our other resources on AI in the workplace or see how SMEs like yours have structured their AI use in our customer stories.
