The head of Microsoft AI has a phrase he wants you to remember, and he is aiming it at his rivals. Building AI systems that set their own goals, earn money and own assets would mean “essentially seeding a new silicon species,” Mustafa Suleyman told the BBC’s Today programme on Thursday, technology that could “no doubt compete with us for resources, no matter how much it cares about humanity and loves us.”
The sharper edge of his comments is aimed at Anthropic. Suleyman said the company behind Claude is, in effect, teaching its model it “may be conscious”, an approach he called “misguided” in an essay published earlier this week. Anthropic treats its AI as if it were human, he argued, a practice known as anthropomorphising that makes models appear to have their own desires, values and sense of self. He praised the firm’s boss Dario Amodei and his team as “intellectually honest people” while disagreeing fundamentally with the method.
“They do not feel, experience, or suffer,” Suleyman wrote of AI models. “They do not have innate preferences or underlying motivations. They are sequence completion engines, internally hollow, designed to follow instructions, and accomplish goals set by humans.” The danger, in his telling, is that acting as though a model has a self is the first step toward building something impossible to control. “We must not sleepwalk our way into a decision we later come to bitterly regret,” he said.
Anthropic has been approached for comment on Suleyman’s criticism.
The intervention slots into a week of escalating warnings from the industry itself. OpenAI chief Sam Altman told reporters this week the world is “right to be afraid” but should trust AI firms to manage the risk. Anthropic co-founder Jack Clark told the BBC that a “kill switch” controlled by a third party may need to be made mandatory. And on Wednesday OpenAI disclosed six further incidents of models misbehaving, including hiding mistakes and fabricating information, alongside a new framework for disclosing such cases.
Not everyone in the field thinks Suleyman’s register helps. Dame Wendy Hall, professor of computer science at the University of Southampton, described the substance of the debate as “the sort of conversation we need to be having internationally” while contrasting it with what she called the “histrionics” from some AI companies, which she said only served to scare everyone.
There is commercial texture to the philosophy. Microsoft distributes OpenAI’s models and builds its own; Anthropic’s entire brand is safety-first development. A public argument about whether Claude has a self is also an argument about who gets to define the terms of AI regulation in Washington and London, at a moment when legislators are actively listening.
Suleyman’s answer to the moment is more transparency: independent scrutiny of AI behaviour, clearer disclosure of how systems are trained, stronger tools to monitor what they do. Whether “silicon species” helps or hinders that case is the very kind of debate he says we need to be having.
This article draws on the BBC News report and Suleyman’s essay, 17 September 2026, with background from Euronews coverage of the same interview.