Microsoft has proposed a code of conduct for its in-house AI models, including expectations that they accept correction, interruption and shutdown. The draft is open for public consultation and could guide model development from 2027, but Microsoft says it is not currently being used to train its models.
Microsoft AI Watch analysis
What happened
The proposal sets out how Microsoft’s future models should respond to human direction, operate within limits and handle instructions from Microsoft, organisations deploying them and individual users. It also says models should not expand their own objectives, resist shutdown or be treated as conscious beings seeking legal rights.
The Daily Star reports that Microsoft AI chief Mustafa Suleyman described the document as a governing framework for future models developed by the company’s AI division. The consultation is due to run for six weeks, after which Microsoft plans to revise the draft. The proposed rules are not yet in use as training guidance; Microsoft expects a revised version to guide development from 2027.
Why it matters
This is a concrete attempt to describe what human control should mean in model behaviour: people should be able to correct, interrupt or shut a system down, and models should stay within limits set by people. That matters as AI systems become more capable and are given more room to act.
A draft code is not the same as a demonstrated safeguard. Its value will depend on what changes after consultation and how the expectations are applied to future models.
Our read
The useful part is the specificity: correction, interruption and shutdown are clearer tests than a general promise to keep AI under control. The harder test comes later, when Microsoft has to show how those principles shape development in practice. For now, this is a proposal to debate, not evidence that the rules already govern its models.
What to watch
- What changes in the draft after the six-week consultation.
- Whether the revised framework appears on Microsoft’s stated 2027 timetable.
- How Microsoft puts the proposed limits into practice during model development.
Discussion spark: Should a company’s voluntary code be enough to define how its AI models respond to correction and shutdown, or should those requirements be independently enforceable?
Sources and evidence
- Microsoft proposes AI code to keep future models under human control – The Daily Star (2 October 2026, 05:49 UTC)
not affiliated with or endorsed by Microsoft