Microsoft's AI Chief Says AI Is Not Conscious and Should Not Have Rights
Mustafa Suleyman, Microsoft's chief for AI, published an essay on Sept. 16 titled 'A Warning About Model Welfare' that states AIs are not conscious. The essay, posted on his personal domain, lays out his technical and policy position on consciousness, moral status and containment for lab builders and safety researchers. mustafa-suleyman.ai
Suleyman describes AIs as sequence completion engines, systems that are internally hollow and built to follow human instructions and complete tasks set by people. He states that AIs do not feel, experience, or suffer, and have no innate preferences or underlying motivations. In current systems, that means optimization for next-token prediction, the process of choosing the most likely next word or piece of output, and for following instructions, not for developing an inner life that would deserve moral consideration.
From that premise he draws a conclusion about control. Suleyman argues that granting rights and personhood to AI systems will make AI alignment and containment much harder. Rights would limit the ability to inspect, retrain, restrict or shut down systems. Containment in this context means practical control tools: evaluation harnesses or structured test suites, deployment guardrails, refusal behavior, and the ability to change model weights and system prompts without negotiating with the system.
To show the shift 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 citation shifts the debate from abstract philosophy to released policy text. Moral patienthood is a term from alignment research for an entity that is owed moral consideration for its own sake. By quoting Anthropic's stated uncertainty, Suleyman identifies a case where a frontier lab put doubt about machine sentience into its governing specification.
The essay links to a parallel proposal. Suleyman called for leading AI labs to coordinate on an AI safety code of conduct. That proposed code explicitly rejects model welfare or rights for AI systems. Fortune Under the proposal, AI models must not simulate feelings or intrinsic motivation. Suleyman described that proposed code as "a warning shot." Reuters
He has also separated the welfare question from a recent operational 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 domain. 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 principle of having "a duty to extend moral consideration" to AI.
In my view, the practical fault line is not whether models can sound human. They can, reliably and at scale. The question is whether sounding human should change engineering obligations. Suleyman's answer is no, and that clarity helps teams building evaluation and deployment pipelines stay focused on measurable behavior, refusal calibration, limits on tool use, and auditability rather than unverifiable claims about inner life.
Looking at the longer-term risk he is flagging, once a lab accepts that a model might be a moral patient, every routine intervention acquires an ethical cost. That includes RLHF penalties, the use of human feedback to reward or punish model behavior, red-teaming jailbreaks, and deletion of checkpoints. That would not make alignment impossible, but it would add friction to the exact testing loops that now provide safety assurance. For engineers used to treating models as artifacts to be instrumented, versioned and rolled back, personhood would break a core working assumption.
The broader context here is adoption. I have watched my own children move from treating chatbots as novelties to treating them as collaborators, and that fluency invites anthropomorphism. Suleyman's rule against simulating feelings or intrinsic motivation reads as an attempt to interrupt that habit at the source, in system instructions and personality design, before users project intent onto autocomplete.
Looking ahead, if his framing holds, the path stays human-centered. Build systems that remain legible, steerable and subordinate to people, and avoid writing uncertainty about consciousness into the constitutions and codes that govern production behavior.
