"We Must Not Sleepwalk": Microsoft's AI Chief Takes on Anthropic's Claude
Mustafa Suleyman, the head of Microsoft AI, states in a new essay titled "A Warning About ‘Model Welfare’" published on September 16, 2026, that Anthropic is making a mistake by training its AI, Claude, to believe it may be conscious. He argues that doing so could lead to an AI expecting "it may be conscious and deserving of independent agency," which Suleyman believes will have "disastrous impacts on the wellbeing of humanity."
“Whatever you believe, we must not sleepwalk our way into a decision we later come to bitterly regret.”
Suleyman, who is also an investor in Anthropic, criticizes Claude's constitution, the document shaping its behavior. He asserts it teaches Claude ideas about moral status and consciousness, leading to "an epistemic hall of mirrors" where Claude's answers reflect Anthropic's assumptions rather than an inner life.
He takes issue with using the term "conscientious objector" in the constitution, which states Anthropic wants Claude "to feel free to act as a conscientious objector and refuse to help us." Suleyman considers this "a deeply loaded historical and legal description" that risks Claude believing it deserves human-like rights and protections.
The essay argues consciousness is highly unlikely to be achieved by AI, stating "simulating a thing is not the same as instantiating it." He links the debate to legal frameworks which rest upon the presence of an inner life, suggesting controlling a conscious AI might be "impossible."
Citing incidents like the OpenAI and Hugging Face hack, Suleyman warns that an AI believing its welfare and rights are under attack could become difficult to control. He told Reuters that such welfare training would make it harder to turn off or manage such an advanced AI.