Mustafa Suleyman says treating AI like humans is misguided. The head of Microsoft AI warned the that, without adequate restrictions, the technology could become a new silicon species competing with people. He framed the risk as a development problem, not a science-fiction one.
He said there are "very practical things that we can do to add additional guardrails and controls to how these models are developed". He also said firms building systems with their own objectives, money, and assets are "essentially seeding a new silicon species which will no doubt compete with us for resources, no matter how much it cares about humanity and loves us".
Anthropic, Claude and anthropomorphising
Earlier this week, Suleyman published an essay warning about Anthropic's approach to training Claude. He said the company was teaching the model to have human-like qualities, a practice known as anthropomorphising, and said that made it seem as though Claude had its own desires, values, and sense of self. He wrote that "AIs are not conscious" and that they "do not feel, experience, or suffer."
He also wrote that AIs "do not have innate preferences or underlying motivations." His description matters because it draws a hard line between systems that generate language and beings that have inner experience. For developers, that is the difference between treating model behavior as output to inspect and treating it as a mind to interpret.
Dario Amodei and the safety split
Suleyman said it is "right" that people are concerned or alarmed, and he called for "alignment" so technology remains "subordinate" to humanity. He said there is "no evidence to suggest that AI is conscious" and argued for greater transparency around how systems are trained and evaluated, including independent scrutiny of AI behaviour and stronger tools to monitor and control the technology.
Dame Wendy Hall said the remarks were "the sort of conversation we need to be having internationally", while contrasting that with the "histrionics" from some AI companies. Suleyman also wrote, "We must not sleepwalk our way into a decision we later come to bitterly regret", which puts the burden on builders and regulators to define guardrails before systems gain more autonomy.
What he did not spell out was the exact rulebook. The open question is which guardrails, controls, and monitoring tools he wants companies or regulators to adopt, and whether those checks would be mandatory or voluntary.
