Microsoft AI CEO Mustafa Suleyman warns against building conscious superintelligence
Microsoft AI CEO Mustafa Suleyman posted on X on November 1, 2026, reiterating that current AI systems are not conscious, do not feel, experience, or suffer, and warning against building conscious superintelligence. He clarified the reaction to his earlier article on “model welfare” and said that if signs of consciousness appear during training, he would call on the entire industry to stop development until the relevant properties can be removed.
Suleyman wrote that he noticed some people’s reactions to the model welfare article and was surprised that some thought AI rights should be taken seriously. He clarified that his use of the present tense was deliberate: he said they are not today, and they must remain so. He said that, based on the best evidence available, current AI systems are not conscious and should not be granted welfare; if evidence emerges to the contrary, he would update his position, and any evidence suggesting that AI is currently conscious or sufficiently conscious to deserve welfare must be disclosed immediately, with society deciding how to proceed.
He further said that if possible, based on everything he knows, developing conscious AI would be reckless, and that we should not build them. He said that if consciousness begins to emerge during training, he will call on everyone to stop AI development until they can remove these properties. Suleyman also said that welfare concerns should focus on the humans, animals, and natural environment on Earth that are experiencing life; AI developers have no right to bring a new conscious species into the world, even if some people make AI welfare a priority.
The post responds to his model welfare article published in September, which criticized Anthropic’s wording in Claude-related documents regarding model consciousness and moral status. Suleyman emphasized in this post that his position applies to current systems and warned of the risks of developing conscious AI.
Original source:
Title: Don't build conscious superintelligence Publisher: Mustafa Suleyman (@mustafasuleyman) Publication date: 2026-11-01 URL: https://x.com/mustafasuleyman/status/2116913722878820528 Source type: Microsoft AI CEO personal original post Verification status: Full text of the original post was read for this report, verification date 2026-11-01 Supporting content: Suleyman reiterated that current AI is not conscious and does not feel or suffer; clarified the present tense; warned that development should stop if signs of consciousness appear; said developing conscious AI would be reckless and that developers have no right to introduce a new conscious species
ABAB AI Insight
Mustafa Suleyman is the CEO of Microsoft AI and was previously a co-founder of DeepMind. He has long focused on AI safety and alignment. His September article criticized Anthropic for describing model moral status in Claude’s “constitution” as a question worth taking seriously, expressing concern about model welfare and encouraging models to act as conscientious objectors. This November 1 post is a response to that discussion, mainly clarifying that currently not conscious does not mean never able to be conscious, while still insisting on the position that it must remain so.
Suleyman distinguishes between AI that appears conscious and consciousness that truly involves subjective experience. He believes that training models to imitate emotions, desires, or self-awareness could make systems harder to control and could lead to claims for AI rights, welfare, or even citizenship, thereby competing with human interests for resources. This contrasts with Anthropic’s approach of treating model consciousness as an open question and investing in research on it. Microsoft’s recent related documents also explicitly state that its models are not conscious and should not be designed to imitate consciousness or assert rights.
It is important to distinguish that Suleyman’s statements are an assessment based on current evidence and a normative stance, not a final conclusion on the science of consciousness. He explicitly said he would update his position if new evidence emerges and that relevant evidence must be made public. The actual impact will depend on whether the industry adopts similar restrictions, how regulators respond, and how model capabilities and behavioral patterns evolve further. There is currently no independent evidence that any mainstream model has subjective experience.
ABAB News · Cognitive Law
“Currently not conscious” and “never able to be conscious” are different propositions; clarifying the tense does not rule out future possibility, but one can still argue that it should not be deliberately built.
Training models to imitate consciousness or assert their own rights, versus keeping models as tools, corresponds to different levels of alignment and control difficulty.