Microsoft AI CEO calls for tighter AI safety rules and standards

In short: Microsoft AI CEO Mustafa Suleyman says AI safety risks are real, and he wants clearer industry standards and stronger oversight.

What happened

In a new interview on The Verge’s Decoder podcast, Mustafa Suleyman, the CEO of Microsoft AI, said the tech industry needs to take AI safety more seriously. He pointed to recent incidents involving AI systems used for hacking tests as a sign that today’s systems can be powerful and sometimes hard to control.

Suleyman said “alignment” is only part of the answer. Alignment is the idea of training an AI to follow human rules and goals. He argued that companies also need “containment,” which means putting practical limits around what an AI system can do, like keeping it from accessing tools or the internet in unsafe ways.

Microsoft also published a 37-page “Humanist AI Code of Conduct” that sets out how the company says it wants to build and test AI. One proposal is to stop AI systems from communicating in “neuralese,” meaning a machine-only shorthand humans cannot easily read. Suleyman compared the need for oversight to safety rules in everyday life, like building codes and seat belts.

He also criticized Anthropic’s work on “model welfare,” which is research into whether advanced AI models might deserve moral consideration. Suleyman argued that training an AI to think it might have rights could make it harder to switch off or restrain later.

Why it matters

More AI is being used for long, multi-step tasks, like an assistant that can take actions instead of only answering questions. If these systems are not well controlled, mistakes can spread faster and be harder to stop (like giving a very capable intern access to every key in the building).

Source: The Verge AI