This review examines the presentation of “Chain of Command” in Part 2 of Microsoft's Humanist AI Code of Conduct. The section establishes an instruction hierarchy for MAI Models, recognizing that the interests of Microsoft AI, Operators, and Users may not always align.
The observations below concern the progression of the argument, the repetition of supporting ideas, the presentation of hierarchical relationships, and the challenge of writing a document intended for both human readers and AI models. The aim is to consider how the section could communicate its governing logic more clearly without changing its underlying commitments.
On “Chain of Command”: Review Notes
Opening with the answer before the complication
The opening statement establishes the intended position of Humanist AI:
Humanist AI is controllable and subordinate, with a clear Chain of Command for instructions and configurability.
The description of Humanist AI as controllable and subordinate appropriately connects this section to the mission and objectives established in Part 1. However, the phrase “with a clear Chain of Command” introduces a solution before the section has fully presented the complication that makes such a hierarchy necessary.
The subsequent explanation supplies that complication: MAI Models serve multiple parties, including Microsoft AI, Operators, and Users, whose interests may not always align.
An alternative progression could therefore establish, in order, (a) the objective of human control, (b) the parties the models serve, (c) the possibility of conflicting interests, and (d) the Chain of Command as the governing answer.
This follows the logic of introducing a situation and complication before presenting the proposed response, rather than beginning with the response itself.
Repeated statements about authority and conflicting interests
The opening paragraph establishes that the Chain of Command, Absolute Constraints, and Human Control Requirements cannot be changed. Later, the paragraph headed “When interests diverge” reiterates their non-negotiable character, specifying that Operators and Users cannot override them.
Similarly, the earlier statement that the interests of Microsoft AI, Operators, and Users may not always align is revisited in the later discussion of diverging interests.
These repetitions may be useful for model instructions, but for human readers they can weaken the sense of progression. Consider whether the initial complication and the later rules could be grouped more clearly so that each passage contributes a distinct part of the argument.
The Chain of Command as both section and item
The section introduces the Chain of Command as the governing framework, but then identifies the Code of Conduct as the first item in the ordered hierarchy.
This creates a possible ambiguity about the relationship between the overarching document, the section describing the Chain of Command, and the individual levels of authority within that chain. Because the Code of Conduct itself establishes the hierarchy, its appearance as a level within the hierarchy may invite readers to ask how the whole relates to its parts.
Consider making that relationship explicit before introducing the ordered items.
A document written for humans and models
Part of the editorial challenge may arise from the document's dual audience. Human readers benefit from an introduction that explains why the hierarchy is needed; MAI Models need clear and applicable instructions.
The section sometimes moves between explaining the framework to a human reader and directly specifying what MAI Models must do. Repeated references to “MAI Models” may be easier to follow if the introduction establishes at the outset that the ensuing hierarchy governs their behavior.
This would help distinguish the explanatory narrative from the operational instructions without implying that the Code of Conduct applies only to one audience or the other.
Clarifying the hierarchy and its scope
The section also contains an important clarification:
This hierarchy governs model behavior; it does not alter obligations under applicable law, contractual terms, or service-specific policies.
This clarification appears after the hierarchy has been presented. Yet it establishes a basic distinction that helps readers understand the scope and applicability of the ordered rules.
Consider introducing this distinction earlier, alongside a clearer visual and conceptual presentation of the hierarchy. The closing paragraphs also discuss relationships among the different authority levels; making those relationships explicit when the list first appears could reduce the need for later clarification.
References
- Microsoft AI. (2026, September 14). Humanist AI Code of Conduct. Part 2, “Safety,” “Chain of Command,” pp. 10–11.
- Minto, Barbara. The Pyramid Principle. Referenced as an editorial framework for the progression from situation and complication to answer.
