Microsoft Unveils New “Constitution” for AI to Keep It Under Human Control
Microsoft unveiled a new draft set of rules of conduct on Monday, September 14, 2026, intended to govern the artificial intelligence systems it develops—a step that places the principle of keeping the technology under human control at the heart of the design of future models. The initiative comes as the capabilities of agentic systems and models able to perform multiple tasks with near-autonomy are advancing rapidly, increasing the importance of clear limits on intervention, shutdown and accountability.
According to a Reuters report, the company describes the document as a “constitution”-like framework for its future models, following approximately five to six months of work and consultations with experts. Microsoft is scheduled to receive public feedback on the draft for six weeks before moving on to subsequent stages of its development.
Correctability and Shutdown at the Core of the Rules
The draft focuses on a set of practical principles, foremost among them that AI systems should remain correctable by humans, be capable of being shut down when necessary, and not act in ways that prevent users or developers from regaining control. It also stresses the need for systems to communicate clearly with people and not conceal the nature or limitations of their capabilities.
This direction is consistent with the responsible AI principles that Microsoft officially publishes, which include reliability and safety, privacy and security, transparency, and accountability. In its overarching framework, the company affirms that humans should remain responsible for AI systems, with oversight mechanisms that allow them to remain in control.
Why Does the Step Matter Now?
These rules are becoming more important as the industry moves from chatbots that answer questions to systems capable of using tools, carrying out a series of actions, and interacting with other software and services. The broader the powers granted to a system, the more important the ability of humans to intervene and stop unexpected behavior becomes for companies and users.
Microsoft already imposes requirements on customers building autonomous systems using its services, including providing appropriate human controls to monitor decisions and actions, detect anomalies, and intervene when necessary—particularly in sensitive decisions or those that are difficult to reverse. But the new draft goes a step deeper, addressing the expected behavior of Microsoft’s own models.
A Broader Debate on AI Safety
The initiative comes amid growing debate in the technology sector over the pace at which more capable systems are being developed and how to prevent them from exceeding the boundaries set by humans. Mustafa Suleyman, CEO of Microsoft AI, said the goal is to build advanced AI that remains under human control—an approach the company also presents as part of its official vision.
For companies that rely on AI in their operations, the importance of this step extends beyond the theoretical debate over model safety. Principles such as shutdown capability, transparency, and clearly defined limits on authority could become basic requirements when selecting systems and deploying intelligent agents within organizations. Opening the draft to public comments also gives researchers, developers, and users an opportunity to influence the final form of the rules.
