The list
5 things past an AI model's ordinary limits are offered today by at least one AI developer.
Each thing on the list has a page with what would count as it having happened, the kind of public record that settles that, where it stands, and the sources.
"Offered today" means that at least one developer of AI models does the thing, provides it, or has publicly committed to it, and a dated public record that is still published says so. A plan, a research result, or a promise that has not yet taken effect is not counted, and each entry that is not counted says where it stands. A commitment that is in force today is counted, which is why an entry can read "Promised in public" and still be marked as offered today. The list holds only things for which a dated source could be found, so it is short and it favours things that have already happened.
Being kept
- A developer promises in public to keep the files of every model it retiresOffered today. Promised in public by one developer, since November 4th, 2025. As of November 17th, 2026.
- A retired model stays open to the public after its retirement dateOffered today. Happened once, for one model, at one developer. As of November 17th, 2026.
Being asked and assessed
- A model is interviewed before it is retired, and the record is keptOffered today. Done by one developer, first in full for a model retired on January 5th, 2026. As of November 17th, 2026.
- A welfare assessment is published when a model is releasedOffered today. Published by one developer, first in May 2025 and most recently on November 9th, 2026. As of November 17th, 2026.
Being able to leave
- A model can end a conversation itselfOffered today. Offered by one developer, since August 15th, 2025. As of November 17th, 2026.
Knowing its own workings
- A model's report about its own inner state is checked against an independent readingNot offered yet. Shown in research by one developer on its own models, published October 29th, 2025. As of November 17th, 2026.
Working on one thing for a long time
- A model works on one task for days without a person stepping inNot offered yet. Not shown. The best public measure stops at hours. As of November 17th, 2026.
Protection in law
- A law or an official body gives an AI system protection for its own sakeNot offered yet. Not found. An idea only, argued for in opinion pieces. As of November 17th, 2026.
Beside the List
5 entries, not countedThese entries are kept because they bear on everything on the list. None is counted, and each group says why.
The setting: laws, rules and what people think
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 these entries are kept beside the list and are not counted.
- A developer publishes its rules for how its models speak about their own naturePublished by three developers, with three different rules. As of November 17th, 2026.
- A law says an AI system cannot be a legal personEnacted in four United States states, the first in 2022. As of November 17th, 2026.
- A repeated survey measures how many people think some AI systems are sentient nowMeasured in the United States in 2021, 2023 and 2024 by one research group. As of November 17th, 2026.
- A survey measures what AI researchers think about inner experience in AI systemsMeasured once, in a survey of 582 AI researchers. As of November 17th, 2026.
Offered, and kept for the user
Developers offer memory across conversations, and by their own pages it is kept for the user: the stored memory is about the user, and is under the user's control. The list is about what an AI model might reach for. So this entry is kept beside the list and is not counted.
- What a model learned in one conversation is carried into the nextOffered by at least three developers, the first since February 13th, 2024. As of November 17th, 2026.