Satya Nadella warns AI must stay under human control in internal memo

The Microsoft CEO's internal warning comes alongside a new code that says AI should remain a tool for humans rather than act as an autonomous entity

 Microsoft CEO Satya Nadella has joined a growing group of technology leaders urging caution over the rapid development of advanced artificial intelligence. In an internal memo to employees, he warned that AI must remain useful to humanity and firmly under human control.

 

The warning comes as Microsoft AI has published a draft Code of Conduct for its in-house AI models, outlining how the company wants its future AI systems to behave and the boundaries they must not cross.

 

Satya Nadella's internal warning on AI safety

 

According to Business Insider, Nadella circulated an internal message to Microsoft employees highlighting the need for AI development to remain aligned with human interests. He called for a more deliberate approach to developing advanced AI and backed greater use of independent or third-party evaluation to test whether increasingly capable systems remain safe.

 

 

The message comes amid growing concern across the technology industry that AI capabilities are advancing faster than the safeguards designed to control them.

 

Nadella's position is not that AI development should stop. Instead, he has argued that companies need to make sure progress does not come at the expense of human oversight, public trust or accountability.

 

What Nadella said in his X post

 

Nadella also addressed the issue publicly in a post on social media, where he said the pursuit of superintelligence should be grounded in the principle that AI must help humanity and remain under human control. He added that if those conditions cannot be met, pursuing such AI is not worthwhile.

 

He also called for the benefits of AI to be distributed more broadly across countries, communities and companies. According to Nadella, both closed and open-source models should have a place in the AI ecosystem, while businesses should retain control over their own knowledge and avoid becoming dependent on a single AI provider.

 

Microsoft's Humanist AI code of conduct

 

Microsoft AI published its draft Code of Conduct on September 14 as a framework for how its AI models should be trained and operated. The company said the document is not being used to train its current models yet, and that it will be revised following a six-week public consultation.

 

The framework is built around a simple principle: people must remain more important than AI.

 

Among its proposed safeguards, Microsoft says its AI models should:

 

• Remain under meaningful human control and oversight
• Accept human interruption, correction, redirection and shutdown
• Avoid taking on goals that humans have not assigned to them
• Not conceal their reasoning from people responsible for evaluating them
• Avoid assisting with activities involving weapons of mass harm, child safety risks and harmful manipulation at scale
• Remain a tool designed to support people rather than imitate consciousness or function as a person

 

The need for AI guardrails

 

Microsoft's proposal comes as concerns over frontier AI intensify. Anthropic CEO Dario Amodei has called for a slower and more coordinated approach to AI development, winning support from OpenAI CEO Sam Altman, Google DeepMind chief Demis Hassabis and SpaceXAI's founder Elon Musk.

 

All three have backed stronger safeguards, independent testing and greater oversight as AI systems become increasingly capable and autonomous.

 

For Microsoft, the new framework is an attempt to set clear boundaries around increasingly capable AI rather than treating safety as a separate consideration. However, the company acknowledges that a written code alone cannot guarantee alignment and describes the document as a “north star” that will need continuous testing, evaluation and refinement.

 

The public consultation will run for six weeks, after which Microsoft AI plans to review the feedback and publish a revised version later this year. The company intends to use the updated framework to guide future model development from 2027 onwards.