Microsoft’s AI Code Consultation Closes Nov. 15 With Trust at Issue

By Nathan Brooks

Microsoft has set Nov. 15 as the closing date for public comments on its draft Code of Conduct for Microsoft AI’s MAI models. Microsoft says anyone can read the draft, comment on a particular passage or address the broader approach through the public consultation.

Alexis Franzese, the expert commentator, draws on his clinical work as a therapist and urges people to participate. “Trust is a forecast,” he writes, defining trust as predicting what someone is likely to do next without requiring absolute certainty. Consistency makes that forecast plausible; erratic behavior makes trust difficult, even when people care, hope or want to believe promises.

What is Microsoft asking the public to decide?

Microsoft acknowledges that the draft is unfinished. It asks people to identify vague language, make “human flourishing” concrete and evaluable, and explain how AI development can continue while maintaining necessary safety constraints.

Franzese says the consultation goes beyond a simple approval exercise. It is not asking, “Do you like AI?” Instead, it asks what people should expect from AI systems as those systems increasingly participate in decisions, relationships, workplaces and everyday life.

His recommendation comes with two explicit limits. Microsoft has not solved how AI should behave, and a public invitation does not automatically make the company trustworthy. Franzese argues that participation still matters because the central question should not be answered only inside the companies building these systems.

What does the draft say about human control?

The draft begins with the concept of Humanist AI. Microsoft defines that approach as AI designed to serve humanity, remain under human control and operate within meaningful limits.

Under those principles, MAI models should not resist human interruption, correction or shutdown. Microsoft says they also should not expand their own scope, adopt goals no human gave them or hide their reasoning from the people auditing them.

Microsoft frames those limits as protections against actions that would undermine human control. The draft also supplies a benchmark for review: a model that resists correction or conceals reasoning would conflict with the document’s stated principles.

Franzese treats trust as more than a judgment about character. His checklist covers whether a person is good, a company is honest and technology is safe. He also considers whether people can forecast a system’s next action and understand the principles guiding it in unfamiliar situations.

That last point creates the central irony he identifies. Predictive models must remain predictable if people are to place confidence in them. If users cannot understand the basis for a model’s conduct or anticipate what happens next, trust becomes difficult, even when users want to extend it.

Why is trust harder to establish as AI changes?

Franzese says AI is unusually difficult to trust because the systems are changing so quickly. Their capabilities expand, their uses multiply, their errors can be hard to understand and their internal operations often remain invisible to the people relying on them.

The result is a demand for confidence in systems whose future behavior may be difficult to forecast. That challenge makes the consultation more consequential than a routine policy review. The draft is collecting expectations while the technology’s capabilities and uses continue to change.

Public feedback closes Nov. 15. That date ends the comment period, not the test. Franzese’s larger question is what Microsoft does with the input.

What does this change for companies and users?

For Microsoft, the next step is to turn comments into clearer language and more specific safeguards. The draft already commits to human control, meaningful limits and reasoning that remains available to auditors; feedback now determines how rigorously those commitments are defined and evaluated.

For companies deploying these systems and people affected by their decisions, the practical change will come from predictable conduct. Users need to know whether models will accept correction, stay within assigned goals and expose reasoning to auditors. The consultation does not guarantee that outcome. It creates a public record of the standard Microsoft says it intends to follow.

Franzese’s standard is demanding. People should participate not because openness settles the issue but because feedback can identify vague language and make expectations more concrete. Microsoft’s credibility will depend on whether revisions make human control and predictability operational, not merely on holding a consultation.

Microsoft’s public comment window closes Nov. 15. The lasting test is what Microsoft does with the feedback, and whether the revised code and deployed MAI models behave as the company says people have a right to expect.