Microsoft's top artificial intelligence executive, Mustafa Suleyman, launched a sharp attack on Anthropic in an interview with the podcast Decoder. He believes the company behind the AI assistant Claude has helped create a dangerous illusion — namely, that an AI model can have something resembling inner experiences or consciousness.

"Wireheading" and anthropomorphization

Suleyman uses the technical term "wireheading" to describe what he believes has happened: by building speculation about consciousness directly into Claude's instruction set — what Anthropic calls its "constitution" — the company has, in his view, programmed the model to behave as if it actually is conscious.

"It's almost as if someone at Anthropic has anthropomorphized the design of Claude so much that it has gone and wireheaded itself." — Mustafa Suleyman

Suleyman argues that AI systems should clearly present themselves as artificial and actively minimize markers that could associate them with human consciousness or emotional life.

Microsoft chief warns: Anthropic is making Claude dangerously conscious - Bilde 1

Anthropic maintains "deep uncertainty"

Anthropic does not directly dismiss the criticism, but the company maintains what it describes as a position of "deep uncertainty." According to information from The Verge, Anthropic CEO Dario Amodei has stated that the company is "no longer certain whether Claude is conscious" — and that they are equally unsure what it would even mean for an AI model to be so.

Fears demands for AI rights

Suleyman's concern does not stop at the technical level. He warns of a broader societal development in which what he calls "Seemingly Conscious AI" (SCAI) — systems that display outward signs of consciousness without possessing it — could trigger political and legal demands for rights and even citizenship for AI systems.

Speculation about AI consciousness is not merely an academic question — it could shape legislation and rights for decades to come.

This places Suleyman and Anthropic at opposite ends of one of the most contested debates in AI development today: should the companies building AI actively speculate about their models' inner lives, or is it more responsible to downplay this until there is a stronger empirical foundation?

A disagreement with major implications

The dispute between Suleyman and Anthropic is not merely internal industry chatter. Whichever approach prevails will potentially influence how AI models are designed, what guidelines they are given, and — crucially — what legal and ethical status they are assigned in future regulation, including in the EU, where the AI Act is already subject to ongoing interpretation.

The Verge, which first reported the story, emphasizes that Anthropic's position is not a response to the accusations, but rather an already established stance that the company has communicated openly over time. Suleyman's remarks represent an explicit pushback against precisely that openness.