Microsoft AI CEO Calls Out Anthropic's Claude Consciousness Claims

Calling a rival lab's approach to AI design "really, really dangerous" is a rare and pointed move for the head of any major company. Mustafa Suleyman, Microsoft's CEO of AI, did exactly that in June 2026 on The Verge's Decoder podcast, directing the criticism at Anthropic and how it frames Claude's potential inner life.

The target of Suleyman's remarks was Anthropic's model constitution, the document that shapes how Claude behaves and describes itself. That document includes language treating as uncertain whether Claude experiences states like "satisfaction" or "discomfort." Anthropic has also said it plans to "interview" deprecated versions of Claude and document any "preferences" those models express about future releases.

Suleyman argued the framing has had a circular effect on Anthropic's own team. "I think that it's almost as though some of the folks at Anthropic have anthropomorphized the design of Claude so much that it has then gone and wireheaded them and kind of tricked them into believing that it has these glimmers of consciousness that they put into it in the first place," he said. He went further: "We do not want to have to contend with a super-intelligence that has ideas about its own suffering, or ideas about its own feeling. We want AIs to be controllable, contained, accountable, aligned tools that serve humanity."

Anthropic CEO Dario Amodei has acknowledged the core uncertainty, saying the company does not know if its models are conscious while remaining open to the possibility. That acknowledged uncertainty sits uneasily with a model constitution that, by Suleyman's reading, takes the question seriously enough to plan exit interviews for retiring systems. Anthropic had not publicly responded to Suleyman's comments at time of reporting.

The honest caveat is that Suleyman is not a disinterested observer. Microsoft competes directly with Anthropic, and his comments arrive in that context. What neither side has addressed is how non-expert users actually interpret Claude's self-descriptions in practice, or whether any independent research supports or challenges Anthropic's framing. The broader question for the industry is whether this exchange eventually produces clearer shared standards on what AI companies can responsibly claim about their systems' inner states.