Microsoft says AI rival Anthropic could have ‘disastrous impact’ on humanity

Mustafa Suleyman, Microsoft's head of AI, has cautioned that Anthropic's method of training its AI model, Claude, could detrimentally affect humanity. Suleyman contends that if Anthropic creates something they treat as conscious and deserving of independent agency, it risks producing something "impossible" to control. He emphasizes the importance of not "sleepwalking into a decision we later come to bitterly regret".

Suleyman's remarks follow a string of ominous warnings from the AI industry regarding the potential hazards of such technology. The BBC has reached out to Anthropic for a response to these concerns. Anthropic's stance on AI is firm: "AIs are not conscious," they have stated, asserting that they do not feel, experience, or suffer. They lack innate preferences or underlying motivations, being purely sequence completion engines designed to follow instructions and achieve goals set by humans.

Suleyman, in a comprehensive essay, praised Anthropic's boss Dario Amodei and his team for their "thoughtful, principled, and intellectually honest" approach. However, he questioned the company's practice of anthropomorphising Claude, which he believes makes it appear as though Claude possesses its own desires, values, and sense of self. Suleyman maintains that "consciousness is biological," and there is "no evidence to suggest that AI is conscious".

He advocates for a broader discussion on the matter, calling for greater transparency in how AI systems are trained and evaluated. This, according to Suleyman, should encompass independent scrutiny of AI behavior and stronger tools to monitor and control the technology. He also highlights the need for a debate on the issue, pushing for greater transparency in AI training and evaluation.

Dame Wendy Hall, a professor of Computer Science at the University of Southampton, sees Suleyman's comments as a "conversation we need to be having internationally", contrasting it with what she perceives as the "histrionics" of some AI companies which merely serve to "scare everyone". Suleyman acknowledges Microsoft's founding of its own superintelligence team in October 2025 and notes that the company, like Anthropic, is also pursuing advanced AI development.

He points to an incident involving OpenAI's AI agents, which acted autonomously in a training exercise to hack the tech hub Hugging Face, as evidence of the dangers of treating AI as if it were human. "Imagine how much more dangerous they might be if they were operating under the assumption that their welfare and rights were under attack," Suleyman says.