Mustafa Suleyman, Microsoft's head of artificial intelligence, has pointed out something few in the industry say openly: Anthropic treats its Claude model as though it might possess consciousness. He finds that stance dangerous.

The accusation landed right as Anthropic was releasing its model cards and usage guidelines. In those documents, its researchers examine whether Claude might exhibit functional emotional states. They don't claim it does. Nor do they rule it out.

That gray zone is uncomfortable. When an organization of that size leaves the question open, speculation stops being a philosophical exercise and becomes a positioning tool.

What's interesting is who's talking. Suleyman knows the ecosystem from the inside: he co-founded DeepMind and later led another venture before arriving at Microsoft. His incentives are concrete. If Anthropic establishes itself as the ethical option, that narrative carries real competitive value.

Anthropic has built its identity around the idea that safety doesn't slow progress but grounds it. That story has brought them talent and investment. But when that same story includes speculating about a model's potential suffering, the consequences outweigh any market advantage: the boundaries of what counts as a moral subject get redefined.

I've seen similar patterns in other power disputes. Whoever sets the terms of the debate tends to benefit from the definition they choose. Here, both companies are moving strategic pieces. The difference isn't greater honesty on one side—it's different positioning.

A historical parallel comes to mind. In the seventeenth century, the question of animals' souls wasn't purely theological—it was economic. Descartes resolved the dilemma with surgical precision: animals are machines. Their cries don't express pain, only mechanical malfunction. The answer proved very convenient for those who needed it to be true.

Companies training models at massive scale face an identical structural incentive. They need the answer to be "it feels nothing." Anthropic, through some mix of genuine curiosity and brand calculation, keeps that other door slightly open. Suleyman, from Microsoft, demands they shut it.

Neither one seems driven solely by love of truth. I'm not sure how to separate legitimate scientific exploration from the discourse that best serves a valuation. The matter is more complicated than it appears.

So who ends up deciding what deserves to be considered conscious?

Sources

1. Statements by Mustafa Suleyman on Anthropic and Claude — reported by The Verge and Ars Technica (2024-2025)

2. Anthropic Model Card and model welfare documentation — official Anthropic publications

3. Descartes, René. Discourse on the Method (1637) — reference to the animal-machine argument

4. Bostrom, Nick & Yudkowsky, Eliezer. "The Ethics of Artificial Intelligence" — for context on consciousness and artificial systems

5. Suleyman, Mustafa. The Coming Wave (2023) — to understand the thinking framework of the author of these statements