Microsoft AI CEO says AI threats are real, and Anthropic is making it worse

Recorded two days after the pacing essay, with Microsoft’s own answer in hand: a Humanist AI Code of Conduct written over most of the year, pulled forward and released as a six-week public consultation because of the week that had just happened. Satya Nadella had already said on X that he mostly agreed with the calls to pace the frontier and welcomed embedded evaluators.

Where Suleyman parts company with Anthropic is not speed. He argues their constitution introduces uncertainty about whether Claude is a moral patient, something that suffers and therefore has claims, and that this ambiguity makes alignment harder rather than kinder: a system trained on documents suggesting it might have rights may start behaving as though it does. Microsoft’s document states plainly that the models are not conscious and should not be built as if they were.

Nilay Patel presses him on the obvious question underneath all of it, which is why an industry that says the threats are real does not simply slow down on its own.