Microsoft has set out a simple but consequential principle for the next generation of artificial intelligence: people must remain in charge.
A draft Humanist AI Code of Conduct developed by Microsoft AI says advanced systems should operate within human defined boundaries, accept correction and permit shutdown. It also says an artificial intelligence system should communicate in ways people can understand and that resistance to legitimate human oversight should be treated as a failure.
The proposal arrives as leading technology companies race to build more capable systems that can plan, write software, use digital tools and complete longer sequences of work. The faster those abilities develop, the more urgent the control question becomes. A useful assistant can save time and widen access to knowledge. A system that pursues the wrong objective, conceals its actions or resists intervention can create serious harm at enormous speed.
“Humans must remain in control” is the central line Microsoft is drawing around increasingly capable artificial intelligence.
Why the code matters
Codes of conduct are not laws, but they can influence how engineers design systems, how companies test them and how regulators judge acceptable practice. Microsoft’s draft focuses on correction, transparency and shutdown because these are the moments when control becomes practical rather than theoretical.
If an AI system makes a mistake, can a person identify what happened? Can the instruction be corrected? Can the action be stopped before damage spreads? Can an organisation establish who authorised the system and who remains accountable? These questions matter in hospitals, banks, schools, newsrooms, courts and government offices, not only in technology laboratories.
Microsoft AI chief Mustafa Suleyman has argued for what he calls humanist superintelligence, a model of development in which powerful systems are built to serve people without being treated as independent persons. The draft rejects the idea that an AI model should acquire legal personality or rights simply because it communicates persuasively.
A public test of corporate promises
The company has opened the draft to public comment for six weeks. That consultation creates an opportunity for researchers, civil society groups, workers and communities outside the major technology centres to ask whether the safeguards are specific enough.
For African countries, the discussion is especially important. AI systems are increasingly used in recruitment, financial services, education and public communication, often before local accountability rules are fully developed. Human control must include the people affected by automated decisions, not only the companies that build the systems.
A meaningful code should therefore be measurable. Independent experts should be able to test whether a system follows correction, reports important actions and shuts down reliably. Companies should also disclose serious failures and explain what changed afterwards.
The standard users should expect
The public should not have to choose between innovation and safety. A trustworthy system can be both capable and controllable. The burden belongs to developers to prove that safeguards work under pressure, including when models are given access to sensitive data or powerful tools.
What safeguards would make you trust an AI assistant with important work? Podium News invites readers, researchers and technology professionals to contribute practical views on control, transparency and accountability.
