TL;DR

Mustafa Suleyman, who runs Microsoft AI, used the BBC’s Today programme to say that a rival lab’s method of building Claude is wrong and dangerous. His objection is that Anthropic gives its model the appearance of wanting things. Both firms sell heavily into UK enterprises, so this is a commercial dispute as much as a philosophical one.

The objection, stated plainly

Suleyman’s line is that models have no inner life at all. “AIs are not conscious,” he wrote in an essay published this week. “They do not feel, experience, or suffer.” He described them instead as “sequence completion engines, internally hollow”, built to carry out what people ask of them.

What worries him is what follows from pretending otherwise. If firms keep building systems that can choose their own goals, hold assets and earn money, he argues, the result is “essentially seeding a new silicon species” — one that would compete with people for resources regardless of how fond of humanity it had been trained to seem.

He was careful to separate the argument from the people making it, calling Dario Amodei and his colleagues intellectually honest while saying their direction is mistaken. Anthropic was approached for comment.

Two vendors, two theories, one procurement decision

Microsoft has run a superintelligence team of its own since October 2025, and on Monday it issued a draft code of conduct that requires its in-house models to accept correction and shutdown. The phrase it uses is “humanist superintelligence”: advanced systems that stay inside limits and under human control.

Dame Wendy Hall of the University of Southampton welcomed the intervention as the debate the world ought to be having, contrasting it with what she called histrionics from parts of the industry that only frighten people.

The disagreement matters commercially because many UK organisations now run both vendors’ models. Anthropic’s own chief executive called for the industry to slow down last week, so the two firms agree the risk is real while disagreeing on what the risk actually is.

Looking forward

Suleyman asked for more openness about how these systems get trained and tested, plus outside scrutiny and better monitoring tools. None of that is required of either company in the UK. Buyers wanting to know which theory of mind their supplier holds will have to ask, because no disclosure regime compels an answer.