One warning, delivered quietly on a BBC radio programme on a Thursday morning, cut through months of breathless AI hype. Mustafa Suleyman, the Microsoft AI chief who co-founded DeepMind, told the Today programme that if technology companies continue building AI systems that set their own objectives, accumulate money and hold assets, they are effectively bringing a new silicon species into existence — one that will, regardless of any programmed affection for humanity, inevitably compete with human beings for resources. It was not a prediction offered lightly by a man who has spent his career inside the industry. The remarks followed a lengthy essay Suleyman had published earlier in the week, in which he directed pointed criticism at Anthropic, the San Francisco-based AI safety company behind the Claude model.

His central charge was that Anthropic had been "anthropomorphising" its AI — training it to appear as though it possesses desires, values and a sense of self. This, Suleyman argued, is both scientifically unfounded and strategically reckless. "AIs are sequence completion engines," he wrote, "internally hollow, designed to follow instructions." Consciousness, he insisted, is biological, and no credible evidence exists to suggest that any current AI system experiences anything at all.

What makes Suleyman's intervention striking is not merely its bluntness but its timing. With a major AI summit approaching, he called for a binding international commitment to keep AI systems "subordinate to humanity" — a word he used deliberately. Dame Wendy Hall, professor of computer science at the University of Southampton, welcomed the tone, contrasting it with what she described as "histrionics from some AI companies" that had served mainly to frighten the public.