Microsoft has published a draft code of conduct for some of its AI products, with a final version planned for 2027. The principles include keeping AI under human control, preventing it from hiding actions and ensuring people can shut it down. The harder question is how those principles will be enforced, and which parts of Microsoft’s AI business they cover.
Microsoft AI Watch analysis
What happened
The draft, described by Computerworld, sets out high-level rules for Microsoft’s own AI models and Copilot. Among them: models should not set their own goals, evade oversight or be used for purposes such as cyberattacks or making deepfakes without consent. The code also rejects treating AI as a person with human rights and calls for a means for people to switch systems off.
The article says Microsoft plans to gather feedback before issuing a final version and putting the code into effect in 2027. It also says Microsoft Foundry, which offers third-party models and tools through Azure, sits outside the Microsoft AI division covered by the draft.
Why it matters
These principles touch the central question of AI governance: what companies promise, and what they actually require of the systems they build and sell. Scope matters too. A code covering Microsoft’s own models and Copilot would not, on the account in Computerworld’s analysis, automatically cover every model available through Microsoft’s cloud.
The draft’s existence is concrete; its practical force is not yet clear. Without detail on enforcement, testing or consequences for breaking the rules, readers are left to judge a set of intentions rather than a working accountability system.
Our read
The principles are a sensible starting point, but “humans remain in control” needs more than a good sentence and a future date. The useful test is whether Microsoft turns the draft into specific requirements, applies them clearly across its AI services, and explains how compliance will be checked. A code of conduct should be a rulebook, not just a well-formatted aspiration.
What to watch
- What changes between the draft and the final code.
- How Microsoft defines oversight, shutdown and prohibited uses in practice.
- Whether the final rules apply to Foundry or other parts of Microsoft’s AI business.
- What testing, reporting or enforcement will accompany the 2027 rollout.
Discussion spark: Should a company’s AI code of conduct cover every model it sells through its cloud, including third-party systems, or only the products it builds itself?
Sources and evidence
- Is Microsoft truly serious about confronting AI’s dangers? – Computerworld (29 September 2026, 11:47 UTC)
not affiliated with or endorsed by Microsoft