Microsoft has prepared a draft internal constitution that defines the limits within which the artificial intelligence systems it develops must act. The basic principle of the text is humans are more important than artificial intelligence It is based on the idea that In this context, a system that resists correction or closure, behaves incomprehensibly, or tries to take over responsibility from humans is considered a failure. The company plans to use the rules, which it has been working on for about five-six months, as a common ground in product development and security decisions.
The draft wants artificial intelligence to communicate with the user clearly and understandably. In addition, the system must not hide its own limits, indicate points of uncertainty, and comply with the direction change given by the human. According to the approach shared by Microsoft, fix and shutdown commands are not seen as negotiable preferences. Thus, the company stipulates in the text that while more autonomous agents become widespread, the final authority will remain with the human.
Satya Nadella also says that an artificial intelligence that does not help people or does not remain under human control is not worth developing. This statement shows that the company does not limit security only to harmful content filters. Microsoft also considers who decides, how the system can object, and how to stop it when an error is detected, as part of its security design. Therefore, the constitution describes the line of conduct that developers will follow before product features.
However, the draft text alone does not constitute technical assurance. Rules need to be translated into training data, model evaluations, and permission systems within the product. In addition, the models and criteria by which the company will monitor these principles have not yet been explained in full detail. That’s why it’s important for Microsoft to share concrete test results, error examples, and fix processes in the next steps.
Complying with the shutdown command is a basic requirement
The company’s research on working life also provides data that supports human responsibility. In the Work Trend Index research covering 20 thousand employees, 86 percent of the participants state that they see the artificial intelligence output as a starting point, not a final product. Quality control is ranked among the important tasks of people with 50 percent and critical thinking with 46 percent.
These results point to a model in which agents will share tasks with humans rather than taking over completely. The human defines the goal, provides the context, and bears responsibility for the outcome. Artificial intelligence, on the other hand, accelerates areas such as research, drafting and repetitive processes. However, as the system becomes more independent, access permissions and stopping mechanisms become more critical. The practical value of the draft constitution also depends on turning these mechanisms into measurable rules for product teams.
Microsoft is currently presenting the text as a draft and aims to improve it with feedback. This approach makes it clear that the rules are not finalized. However, the collection of principles such as not resisting closure, being understandable and prioritizing human decision in a single document draws an important limit. The question now will be how the company will verify these promises in real products and to what extent it will disclose rule violations to the public.