Mustafa Suleyman Says AI Developers Should Not Build Uncontrollable Systems

Microsoft AI CEO Mustafa Suleyman says developers should not create AI systems they cannot control. In a newly published CNBC Squawk Box interview, he discussed that position alongside AI safety concerns, product liability for AI companies, and Microsoft’s code of conduct for its models.

The interview comes after Microsoft AI published a first draft of its Code of Conduct for MAI models on September 14. Microsoft says the document is meant to guide how its models are developed and how they function during deployment. It is open for public comment for six weeks; Microsoft says it will review feedback, publish a summary, and issue a revised version later this year.

Control is the stated baseline

Microsoft says its MAI models should be designed for people first, under human control, and shaped by human direction. The company presents the code as more than a screen for prohibited prompts: it is intended to influence both the creation of models and their conduct after deployment.

The limits reach beyond controllability

The proposed restrictions also define requests Microsoft’s models should not help fulfill. They include weapons development, procurement of dangerous substances, and violent or sexually explicit content.

Other proposed boundaries

• Models should not be built to imitate consciousness.
• Models should not be entitled to rights.

Those provisions show that Microsoft’s draft takes a position not only on what an AI system may do, but also on how it should be understood. Suleyman has separately described fears of a possible loss of control as real, reinforcing the premise behind the company’s proposed rules.

Suleyman’s CNBC appearance places that principle beside a live discussion of safety and liability. Microsoft is asking the public to weigh in on its approach, while retaining discretion over what feedback reaches the revised code.