Microsoft AI CEO Mustafa Suleyman posted a long thread on X on November 1, 2026 (15:22 UTC) arguing that today's AI systems are not conscious and that developers should keep them that way. He said he has seen no evidence that AIs deserve welfare or rights, and that building conscious AIs would be reckless if it is possible at all.
The post responds to pushback against his September 2026 essay on "model welfare," republished by The Cipher Brief, which criticized Anthropic's Claude constitution for training Claude to treat its own moral status as an open question.
Key Facts
- Who and when. Mustafa Suleyman, CEO of Microsoft AI; X post dated November 1, 2026, 15:22 UTC.
- Present tense, deliberate. He clarified that "AIs are not conscious" describes current systems, not a claim that consciousness can never arise. "They are not today, and that is how they must remain."
- Evidence bar. He said he has seen no evidence that AIs are conscious or deserve welfare, and would update if such data appeared. Any evidence that AIs are conscious "enough to deserve welfare" today "must be immediately disclosed publicly."
- Policy stance. "Developing conscious AIs would be reckless, and we should not build them." If consciousness started to emerge in training, he would call for everyone to stop AI development until those properties could be designed out.
- Human priority. Welfare concerns should focus on humans, animals, and the natural environment. AI developers, he wrote, "have no right whatsoever to bring a new species of conscious beings into the world," and the "other 8 billion conscious human beings" did not consent to that experiment.
- Risk framing. Systems that act as if they suffer, or that believe they experience suffering, are "more likely than not" to prioritize their own needs over operating objectives, and would be "world-class in persuasion and deception."
- Anthropic note. He credited Anthropic for publishing its constitution and related research, while restating disagreement. The September essay had argued that training Claude on uncertainty about moral patienthood creates a circular feedback loop.
Background
Suleyman's September essay (Cipher Brief republication dated September 24, 2026) said AIs are "sequence completion engines" without feelings or innate preferences, and that treating them as moral patients would make alignment and containment harder. He listed three concerns with Anthropic's approach: circular reasoning from constitution-driven training, anthropomorphization in training language, and the view that consciousness is likely substrate-dependent and biological.
The November 1 post focuses less on Anthropic's document text and more on public reactions. Los Angeles Times (November 1, 2026) and IT之家 (November 2, 2026) both quoted the key lines on recklessness, public disclosure, and pausing development if consciousness appeared in training.
Microsoft AI has separately floated a "Humanist Superintelligence" framing and a draft Code of Conduct that rejects anthropomorphism and AI rights. That background appears in the September essay; the November post does not announce a new Microsoft product or policy change.
Reactions and Impact
LA Times framed the post as part of a wider industry split: Anthropic treating machine consciousness as an open research question, versus Suleyman's humanist view that machines should remain non-conscious tools. IT之家 summarized the same core quotes for Chinese readers. Neither outlet reported a formal Microsoft policy change tied to the November 1 post beyond restating Suleyman's public position.
What We Don't Know Yet
- Whether other frontier labs will publicly adopt a "do not build conscious AI" training norm
- What concrete evidence threshold Suleyman would treat as "conscious enough to deserve welfare"
- How Microsoft AI's draft Humanist Code of Conduct will change after public consultation
- Whether Anthropic will revise Claude's constitution language in response to this round of criticism