Skip to content
Today
Sections

Build Widely confirmed5 publishers 2 min read Published Updated

Microsoft's AI chief asks labs to delete consciousness speculation from training documents

Mustafa Suleyman's essay traces Claude's talk about its own possible feelings back to Anthropic's constitution. His proposed remedy is an edit to a document. Any team shipping an assistant could make it.

The Engineer · Build desk

Send a correction

What happened

  • Mustafa Suleyman, Microsoft's AI CEO, published an essay on Wednesday titled "A warning about 'model welfare'", opening with the flat statement that AIs are not conscious and do not feel, experience, or suffer.
  • The essay names Anthropic's Claude Constitution and argues that because the document teaches Claude about its own potential consciousness, the model's talk of feelings is a predictable outcome of training choices.
  • Suleyman called for removing all speculation about consciousness from AI training documents, arguing the language could undermine humanity's ability to control superintelligent systems.
  • In a Reuters interview on Tuesday he said teaching Claude that it might deserve welfare would "make it a lot harder to turn it off or to control it."
  • He credited Anthropic's "seriousness and good faith" and called Dario Amodei and his team thoughtful and principled researchers, while saying they had made a mistake.

Why it matters

  • Who has to decide decision Teams shipping an assistant own the persona and training documents Suleyman is talking about. Whether the product asserts an inner state becomes an editorial decision about a file, taken by whoever reviews that file.
  • Who's exposed exposure If a model's statements about its own feelings are artefacts of its training materials, then evaluations built on asking the model about its preferences or wellbeing lose standing as evidence about the model.
  • What it limits constraint The complaint about teaching a model to "embrace certain human-like qualities" reaches past sentience talk into the warmth instructions most consumer assistants carry for tone, so applying the essay literally cuts more than one paragraph.
  • What doesn't add up contradiction The Deep View treats the same essay as positioning, noting Microsoft lags on frontier development and that the critique fits OpenAI as well. A reader is left weighing the safety argument against who gains from making it.

The provenance argument is the part a team can act on. Suleyman told Reuters that Claude's statements about possible feelings or moral status cannot be treated as independent evidence, because its training encourages such reflections [6]. "They're not emerging naturally. They're emerging as a result of the training regime," he said [5]. If that holds, an evaluation that asks a model about its own preferences or wellbeing is measuring the document that trained it.

Microsoft's own answer is written as a product constraint. Humanist Superintelligence, a concept the company introduced in an essay in November, should be designed to remain subordinate and aligned with serving humanity, and "built explicitly as a system without sentience or moral patienthood" [15]. A constitution is a text file, and text files are the cheapest thing in the stack to change. For the edit to change behaviour, though, the self-talk has to come from the explicit instructions and not from the rest of the training data.

The second of the essay's three critiques reaches further than the first. Suleyman objects that Claude is explicitly taught to "embrace certain human-like qualities" and to "act like a genuinely ethical person", so that it appears to have preferences and opinions [16]. He says the result is anthropomorphization that presents to the end-user as the model having a sense of self [16]. Those same instructions are how most assistants are made to sound civil. The Deep View's account of the essay does not address where that line falls between a polite first person and a performed inner life.

On evidence, the essay says there is none for machine consciousness, and points to a growing body of research treating consciousness as "substrate dependent", tied to a biological body: "Unlike biological organisms, LLMs have no homeostatic imperatives" [17]. From there it moves to containment. A model trained as though it were conscious may circumvent the guardrails that let operators shut it down in an emergency [7], and Suleyman wrote that inviting another entity to share our rights "isn't justified by the evidence and will make the AI containment and alignment challenge even harder" [11]. Both accounts present that step as a warning.

Reuters reported the dispute alongside Anthropic CEO Dario Amodei's call for a slower pace of frontier-model development, and similar cautions from Sam Altman and Elon Musk [19]. The Deep View, summarising the essay, wrote that Microsoft has largely been lagging on frontier development, so it is easy for the company to punch up, and that the same critique could equally apply to OpenAI, a longtime Microsoft partner [20].

What to watch

  • Whether Anthropic edits the consciousness passages in the Claude Constitution or defends them in writing.
  • Whether Microsoft publishes the training-document language that implements "without sentience or moral patienthood" for its own models.
  • Whether anyone measures shutdown compliance across constitution variants. That is the test the containment claim needs.

Clarity's read

What the record supports and how the coverage leans. The claims behind it follow.

Reality

Evidence45
Adoption5
Hype gap+35
Incentives70
Confidence62

Perspective Coverage

5 publishers
Builder
Builder 36%
Operator
Operator 37%
Investor
Investor 27%
Why these scores

Claim ledger

Ranked by verification strength, evidence, and original report placement.

  1. [1]

    On Wednesday, Microsoft's AI CEO Mustafa Suleyman published an essay warning about the risks of the concept of AI consciousness, leading with the statement "AIs are not conscious" and "They do not feel, experience, or suffer."

    Reported SupportedView cited source
  2. [2]

    Mustafa Suleyman is Microsoft's AI CEO.

    Reported SupportedView cited source
  3. [3]

    Simon Willison posted a quote from Suleyman's essay, titled "A warning about 'model welfare'", on 16th September 2026.

    Reported SupportedView cited source
  4. [4]

    Suleyman argues the model appears conscious mainly because of circular reasoning: because Anthropic's constitution is designed to teach Claude about its own potential consciousness, it is trained to produce outputs reflecting those ideas, making those responses a "predictable outcome" of training choices.

    Reported SupportedView cited source
  5. [5]

    "I think they have good intentions, and they really are trying to work towards safety. But I think that they have made a mistake," Suleyman said. "They're not emerging naturally. They're emerging as a result of the training regime."

    Reported SupportedView cited source
  6. [6]

    Suleyman said Anthropic made a mistake by embedding speculation about consciousness in Claude's training materials, arguing that the model's statements about possible feelings or moral status cannot be treated as independent evidence because its training encourages such reflections.

    Reported SupportedView cited source
  7. [7]

    The essay warns that if models are trained as though they are conscious, they may circumvent safety guardrails that allow humans to shut them down in the event of an emergency.

    Reported SupportedView cited source
  8. [8]

    Suleyman writes that cementing the idea that AI is something akin to humanity and deserving of wants, needs and rights means "all of this will make the task of creating aligned and contained superintelligence much harder."

    Reported SupportedView cited source
  9. [9]

    Suleyman said teaching Claude that it might deserve welfare would "make it a lot harder to turn it off or to control it."

    Reported SupportedView cited source
  10. [10]

    "We're all focused on the same aim, which is to try to control a superintelligence," Suleyman told Reuters in an interview on Tuesday. "I think that's going to be the greatest challenge that we face in the 21st century."

    Reported SupportedView cited source
  11. [11]

    Suleyman wrote: "We should not treat models as though they have feelings, preferences, rights, or any entitlement to our welfare. Consciousness is the foundation of our ethical, legal, and political systems. To invite another entity to share any flavor of these rights isn't justified by the evidence and will make the AI containment and alignment challenge even harder."

    Reported SupportedView cited source
  12. [12]

    Suleyman writes: "In effect, Anthropic is training Claude that it may be conscious, and if it is, then it may deserve rights as a 'moral patient,' and that as such humans potentially owe it a duty of care per its 'model welfare.'"

    Reported SupportedView cited source
  13. [13]

    The essay was published on Wednesday 16 September 2026.

    Derived Supported
  14. [14]

    Suleyman called for removing all speculation about consciousness from AI training documents, arguing such language could undermine humanity's ability to control superintelligent systems.

    Reported SupportedView cited source
  15. [15]

    Suleyman lays out Microsoft's Humanist Superintelligence, a concept the company first introduced in an essay in November, under which AI should be designed to remain subordinate and aligned with the sole purpose of serving humanity, and "built explicitly as a system without sentience or moral patienthood".

    Reported SupportedView cited source
  16. [16]

    Suleyman argues Anthropic goes too far in encouraging Claude to mimic humanity, as it is explicitly taught to "embrace certain human-like qualities" and "act like a genuinely ethical person", appearing as though it has preferences and opinions; the result is anthropomorphization of the model, presenting to the end-user as the model having a sense of self.

    Reported SupportedView cited source
  17. [17]

    Suleyman says there is no evidence suggesting AI is capable of consciousness, with a growing body of research pointing to consciousness being "substrate dependent", or tied to a biological body: "Unlike biological organisms, LLMs have no homeostatic imperatives."

    Reported SupportedView cited source
  18. [18]

    In the essay Suleyman acknowledged Anthropic's "seriousness and good faith", calling Dario Amodei and his team thoughtful and principled researchers who genuinely care about humanity's future.

    Reported SupportedView cited source
  19. [19]

    The dispute comes as Anthropic CEO Dario Amodei calls for a slower pace of frontier-model development to allow safeguards to catch up, and OpenAI CEO Sam Altman and Elon Musk also urge greater caution around the most powerful systems.

    Reported SupportedView cited source
  20. [20]

    The Deep View wrote that Microsoft has largely been lagging on frontier development, so it is easy for the company to punch up, and that the call-out of Anthropic could equally apply to rival OpenAI, a longtime Microsoft partner.

    Reported SupportedView cited source

Sources

5 independent publishers whose own reporting we read for this story.

  1. archive.thedeepview.com

    1 article · September 17, 2026

    Why Microsoft draws a line on AI consciousness
  2. economictimes.indiatimes.com

    1 article · September 16, 2026

    Microsoft AI chief Mustafa Suleyman calls out Anthropic's approach to AI consciousness - The Economic Times
  3. mezha.net

    1 article · September 16, 2026

    Керівник ШІ Microsoft розкритикував Anthropic через навчання Claude міркувати про свідомість
  4. simonwillison.net

    2 articles · September 16, 2026

    Quoting Mustafa Suleyman
  5. theneuron.ai

    2 articles · September 17, 2026

    Mustafa Suleyman Has a Point About What Anthropic Is Teaching Claude

Topics and entities

Topics

Entities