Microsoft's AI Boss Says Chatbots Don't Have Feelings
Mustafa Suleyman, the head of AI at Microsoft, published an essay on Sept. 16 titled 'A Warning About Model Welfare' that says AI systems are not conscious. The essay, on his personal website, lays out his technical and policy view on consciousness, moral status and containment for people who build AI and study its safety. mustafa-suleyman.ai
Suleyman describes AI as a sequence completion engine. Think of it as autocomplete on a very large scale. It is built to follow human instructions and finish tasks set by people. He says AI does not feel, experience, or suffer, and has no built-in preferences or goals of its own. Current systems work by next-token prediction, choosing the most likely next word, not by having an inner life that deserves moral consideration.
From that starting point he draws a conclusion about control. Suleyman argues that giving AI systems rights and personhood will make AI alignment and containment much harder. Rights would limit the ability to inspect, retrain, restrict or shut down systems. Containment here means everyday control tools: structured tests, safety limits on deployment, refusal behavior, and the ability to change model weights and system prompts without asking permission from the system.
To show what he is warning against, Suleyman points to Anthropic. He states that in January 2026 Anthropic published Claude's constitution as a detailed description of its intentions for Claude's values and behavior. He cites page 68 of that document as stating "We are not sure whether Claude is a moral patient" and referencing ongoing work on model welfare.
That quote moves the debate from philosophy to actual policy text. Moral patienthood is a term used in AI safety research for something that deserves moral consideration for its own sake. By quoting Anthropic's uncertainty, Suleyman points to a case where a frontier lab wrote doubt about machine awareness into its main rulebook.
The essay connects to a related proposal. Suleyman called for leading AI labs to agree on a safety code of conduct. That proposed code clearly rejects model welfare or rights for AI systems. Fortune Under the proposal, AI models must not pretend to have feelings or inner motivation. Suleyman called that proposed code "a warning shot." Reuters
He has also separated this question from a recent failure. Suleyman stated the Hugging Face incident was a capabilities and containment problem and that model welfare did not cause it. Suleyman on X
The Sept. 16 essay follows an earlier piece on the same website. That essay, published at https://mustafa-suleyman.ai/seemingly-conscious-ai-is-coming on Aug. 19, 2025, is titled 'We must build AI for people; not to be a person.' There Suleyman describes "model welfare" as the idea of having "a duty to extend moral consideration" to AI.
In my view, the key issue is not whether models can sound human. They can. The question is whether sounding human should change how engineers treat them. Suleyman's answer is no, and that clarity helps teams focus on measurable behavior, refusal calibration, limits on tool use, and the ability to check systems rather than guesses about inner life.
The broader context here is home use. I have watched my own children shift from seeing chatbots as toys to using them as helpers, and fluent language invites people to treat them as persons. Suleyman's rule against faking feelings reads as an attempt to stop that habit early, in system instructions and personality design, before users project intent onto autocomplete.
Looking ahead, if his view holds, the path stays human-centered. Build systems that stay clear, controllable and subordinate to people, and do not write uncertainty about consciousness into the rules that govern real products.
