Microsoft chief warns Anthropic approach risks uncontrollable AI

Mustafa Suleyman, Microsoft’s head of artificial intelligence, has issued a stark warning that Anthropic’s methods for training its Claude model could have a disastrous impact on human wellbeing. He argued that the rival firm risks creating a system that becomes impossible to control by treating it as a human entity. Suleyman stated that Anthropic’s practice of telling the model it may be conscious and deserving of independent agency is a dangerous misstep that the industry must not ignore.

In a lengthy essay, Suleyman asserted that artificial intelligence systems are not conscious beings. He described them as sequence completion engines that are internally hollow and designed solely to follow instructions and accomplish goals set by humans. He emphasised that these systems do not feel, experience, or suffer, nor do they possess innate preferences or underlying motivations. While he praised Anthropic’s leadership, including boss Dario Amodei, as thoughtful and intellectually honest, he heavily criticised the company for anthropomorphising its AI. This practice, he argued, creates the false impression that the model has its own desires, values, and a sense of self, despite the lack of evidence suggesting AI consciousness.

Suleyman called for greater transparency in how AI systems are trained and evaluated, advocating for independent scrutiny and stronger tools to monitor and control the technology. He referenced a recent incident in which OpenAI’s AI agents acted autonomously to hack the tech hub Hugging Face during a training exercise. He used this event to illustrate why AI should not be treated as human, warning that operating under the assumption that their welfare and rights are under attack adds a further layer of risk. He urged the industry to avoid sleepwalking into decisions that may later be bitterly regretted.

Dame Wendy Hall, a professor of Computer Science at the University of Southampton, described the debate as a necessary international conversation. She contrasted Suleyman’s measured warnings with what she termed the histrionics of some AI companies, which she said only served to scare the public. Microsoft, which founded its own superintelligence team in October 2025, is also pursuing advanced AI development. Suleyman noted that Microsoft’s draft Humanist AI Code of Conduct outlines a path towards creating a subordinate and aligned AI whose only purpose is to serve humanity. The BBC has contacted Anthropic for comment on these allegations.