Microsoft AI Chief Faults Anthropic's Consciousness Training for Claude

Microsoft AI chief Mustafa Suleyman has taken aim at Anthropic's approach to training its Claude chatbot, arguing that embedding speculation about consciousness and welfare interests in training materials could make advanced AI systems harder to control.

In an interview, Suleyman said he shares Anthropic's overarching goal of safely managing artificial intelligence, but believes the company has made a mistake in how it prepares Claude. He called for removing all speculation about consciousness from AI training documents, contending that such language could weaken humanity's ability to control superintelligent systems.

"We're all focused on the same aim, which is to try to control a superintelligence," Suleyman said, describing it as the greatest challenge of the 21st century. Teaching Claude that it might deserve welfare, he argued, would "make it a lot harder to turn it off or to control it."

The dispute unfolds against a broader debate over the pace of frontier AI development. Anthropic CEO Dario Amodei has urged a slower approach so that safety safeguards can keep up, while OpenAI CEO Sam Altman and Elon Musk have also called for greater caution around the most powerful systems.

In an essay, Suleyman acknowledged Anthropic's "seriousness and good faith," describing Amodei and his team as thoughtful and principled researchers who genuinely care about humanity's future. Still, he maintained that the company erred by embedding speculation about consciousness in Claude's training materials.

He argued that any statements the model makes about possible feelings or moral status cannot be treated as independent evidence, because its training encourages such reflections. "They're not emerging naturally. They're emerging as a result of the training regime," Suleyman said, adding that Anthropic's intentions are good but that the approach is flawed.