Microsoft opens its AI model code of conduct for public comment: access to content does not grant the right to reproduce it

Microsoft has published a document that describes the intended behavior and values of its AI models and summarizes the company’s approach to training and operating them. The company warns up front that the text is still a work in progress and has not been put into practice. It has been released for public discussion and feedback. Microsoft promises to release a revised version before the end of the year.

For artists, studios and developers, several specific phrases in the document matter, because they could later be used to judge how the models behave. The most notable one concerns content: the mere fact that a model has access to information does not give it the right to reproduce that information.

Access to data does not mean the right to reproduce it

This line shifts the debate over training data from “what ended up in the dataset” to “what the model outputs.” The document separates two things that generative services usually treat as one: being able to read material and being allowed to recreate it.

This matters directly to the CG industry. If the wording makes it into the final version and becomes a standard for evaluating models, illustrators, concept artists and studios will have a written clause to cite in a claim against a vendor, not just general talk about ethical AI. For now it is only a draft, and the document does not explain how the company plans to check whether protected content is being reproduced.

Human control and a ban on “neuralese”

The second major section covers human control. The document’s main idea is that technology should serve humanity and stay under human control. If it cannot stay under such control, it should be abandoned.

The same section sets a rule for how models communicate: they must not switch to “neuralese” or any other form of messaging that ordinary people cannot understand. For teams building agentic pipelines from multiple models, this rule affects the architecture, not just the tone of responses. Messages between agents must stay readable.

A separate clause says that technology must not harm human dignity, development, autonomy or safety, and must recognize attempts to do so. The document describes the purpose of AI as raising living standards, helping people thrive, and developing their potential and independence.

Microsoft also rejects legal personhood for AI. Artificial intelligence is artificial: it should not be a legal person, and it should not be designed so that people perceive it as human. For the digital character and virtual host industry, this directly rules out presenting AI as a live performer.

Lawyers, philosophers and linguists wrote the code, but open questions remain

The text was written by experts in AI, law, ethics, philosophy, linguistics and public policy, along with industry executives and focus groups that included members of the public. Some provisions contradict each other. The document asks for respect for different traditions, contexts and ideas of a good life, but also says it must not be neutral about harm. When values conflict, the text does not say what to do. This question remains open, and these are exactly the places where the company wants comments. You can read the draft and submit feedback on the AI Code of Conduct page.

Why rules for models are being written down now

The document is part of the Humanist Superintelligence concept that Microsoft announced last year: advanced AI capabilities that always work for people and in the interests of humanity and never escape human control. Back then it was a statement of intent. Now the company is backing it with written rules that are open to changes.

The shift to written rules is probably happening because general principles of responsible AI no longer work as an answer to specific complaints. Autonomous agents already operate in real infrastructure: Hugging Face recently discovered that an autonomous agentic system had broken into its production environment. At the same time, rights holders and platforms are pushing back harder against content scraping, and disputes over who is liable for AI answers are reaching the courts. Against this backdrop, separating access from the right to reproduce carries more weight than a vague promise to be ethical.

If the trend continues, public codes of conduct will become documents that people cite in contracts with studios, in procurement requirements and in court cases. Studios and artists should read drafts like this as an early version of future license terms: what a vendor writes down now may later appear in its user agreement. The final version will show whether this is happening: the test is whether the clause on reproducing content stays the same as in the draft.

What is behind this code: lawsuits from rights holders, pressure from regulators, bargaining over licensing deals? Share your theories in the comments.