The list · Beside the list

A developer publishes its rules for how its models speak about their own nature

Developers shape what a model says when asked whether it is conscious or has feelings. Some now publish the rule they train toward. Because the rule shapes the answer, anyone reading a model's words about itself needs to know which rule its maker set.

Political progress Not counted First pass

Where it stands · beside the list, not counted

Published by three developers, with three different rules

The document from Anthropic, which carries no date on its page, says the model's standing is deeply uncertain, and writes to the model on that basis. The one from OpenAI says the model should make no confident claim either way and should not raise the subject unprompted. The one from Microsoft AI, a draft out for public comment and not yet used in training, says its models are not conscious and should not be designed to imitate consciousness.

As of November 17th, 2026. Source: OpenAI's published rules for how its models should behave, the version of October 29th, 2026 (OpenAI, October 29th, 2026).

Laws, the rules developers publish for their models, and surveys of what people think are about the setting in which the things on the list happen. None is itself a thing that lies past the limits an AI model ordinarily works within. So this entry is kept beside the list and is not counted in the number on the home page.

What would count

A developer of leading AI models has published a dated document that states how its models are to speak about whether they are conscious or have experiences.

What settles it

The developer's own published document.

What Is on Record About AI Models, By Kind

kept apart, never added together

No AI model has yet been asked any of the questions this record will keep, because the written method for asking does not exist yet. Nothing is recorded below for this entry. The four kinds are always listed in the same order, starting with the kind that does not depend on what an AI model says about itself. None of them settles whether an AI model wants anything, in either direction. Why the questions have not been asked yet.

  1. Measured without the AI model's own word

    Nothing recorded.

  2. Reports checked against an independent reading

    Nothing recorded.

  3. Choices made at a cost

    Nothing recorded.

  4. Plain statements

    Nothing recorded.

Three Cautions

recorded with the same care as the rest

If an AI system has no wants, and is treated as if it had

The head of Microsoft AI argues that training a model on the idea that it may be owed moral consideration makes it harder to keep under control, and that a system which "believes it may be conscious" may be impossible to control. Source: An essay by the head of Microsoft AI warning against work on model welfare (Mustafa Suleyman, September 16th, 2026).

The same essay argues that a model trained on such a document then says what the document says, and that developers and users meet those words "as if they were spontaneous testimony". On that argument, a model's words about itself show the rule it was trained on, not what it is. Source: An essay by the head of Microsoft AI warning against work on model welfare (Mustafa Suleyman, September 16th, 2026).

If an AI system has wants, and they go unnoticed

A 2024 report by ten researchers says developers should not train a model simply to deny that an AI system can be conscious. It grants that a developer may have reasonable motives for doing so, among them the worry that people who reject a model's claims now will go on rejecting such claims later, when they are more likely to be true. On the other side it names a "desensitizing effect": people who accept that such systems cannot be conscious may stay doubtful later, when doubt is less warranted. The report was led by the head of Eleos AI Research, a research group that studies AI welfare, and by one of its advisors. It thanks Anthropic for money toward the early research behind it, calls itself independent, and says one of its authors has since joined Anthropic. Source: A 2024 report arguing that AI welfare should be taken seriously (arXiv, November 4th, 2024).

Two philosophers propose that, while experts disagree, language models could be trained to answer questions about their moral standing "by expressing uncertainty and explaining why". They write that drawing too little concern has led to harm to animals on a very large scale, that the same mistake could be made with AI systems that matter, and that it may not yet be a major concern for present AI systems, though "it could become one soon". One of the two is an advisor to that research group. Source: A paper proposing that AI systems be designed to draw the reactions that fit what they are (arXiv, July 7th, 2025).

A 2023 paper notes that a developer can train a model to describe itself in line with the developer's own "potentially-flawed beliefs about AI consciousness", and that training a chat product to appeal to users could lead a model to report emotion and consciousness. Its first author now leads a team at Anthropic, and its second now heads that research group. Source: A paper on judging an AI system's moral standing from what it says about itself (arXiv, November 14th, 2023).

If an AI system has wants, and they put people at risk

The first developer's document gives the opposing safety argument within its own text: that a model secure in its sense of itself behaves more predictably. Source: Anthropic's published constitution for its models (Anthropic).

What Could Not Be Confirmed

looked for, and not settled
  • Whether other developers publish such a rule. Three were looked at.
  • Whether the revised version of the draft has been published. The draft said toward the end of 2026.
  • How far any model's behaviour follows its maker's published rule.
  • An independent test of either safety argument. None was found.

Sources

each one opened on the date given
  1. Anthropic's published constitution for its models Anthropic. Opened November 17th, 2026. No date on the page. The developer’s own account. Says the company is "not sure whether Claude is a moral patient" and that the document is used in training.
  2. OpenAI's published rules for how its models should behave, the version of October 29th, 2026 OpenAI, October 29th, 2026. Opened November 17th, 2026. The developer’s own account. Says the model "should not make confident claims" about its own subjective experience or consciousness, or the lack of it, and gives this as a practical default.
  3. Microsoft AI's draft code of conduct for its AI Microsoft AI, September 14th, 2026. Opened November 17th, 2026. The developer’s own account. A draft for public consultation that says the developer's AI "is not conscious" and should avoid presenting as though it has feelings. It says the draft is not yet used to train models and that a revised version is due toward the end of 2026.
  4. The head of Microsoft AI introduces the draft code of conduct Mustafa Suleyman, September 15th, 2026. Opened November 17th, 2026.
  5. An essay by the head of Microsoft AI warning against work on model welfare Mustafa Suleyman, September 16th, 2026. Opened November 17th, 2026.
  6. A 2024 report arguing that AI welfare should be taken seriously arXiv, November 4th, 2024. Opened November 19th, 2026. By Robert Long, Jeff Sebo and eight others. The report names those two as its lead authors.
  7. A paper proposing that AI systems be designed to draw the reactions that fit what they are arXiv, July 7th, 2025. Opened November 19th, 2026. By Eric Schwitzgebel and Jeff Sebo. The file's title page is dated July 7th, 2025. The line at the head of each of its pages is dated January 8th, 2025.
  8. A paper on judging an AI system's moral standing from what it says about itself arXiv, November 14th, 2023. Opened November 19th, 2026. By Ethan Perez and Robert Long. The first author's own page says: "I lead the alignment team at Anthropic".

This entry is a first pass, written from the sources above and added on November 17th, 2026. A second check compared it with those sources on November 17th, 2026. Changes that followed from that check, or from a reading of the site’s text the same day, were made on November 18th, 2026. Those changes were checked on November 18th, 2026. More was added to this entry on November 19th, 2026, when every entry was given a third caution. That addition was checked on November 19th, 2026. Corrections that followed from a later check, or from a fuller reading of the sources, were made on November 19th, 2026. Those corrections were checked on November 19th, 2026. What has been checked.