The list · Beside the list

What a model learned in one conversation is carried into the next

By default a chat model starts every conversation knowing nothing of earlier ones. Developers now offer memory: notes or summaries of past conversations are stored and given back to the model later. The stored memory is about the user, and is under the user's control.

Technical progress Not counted First pass

Where it stands · beside the list, not counted

Offered by at least three developers, the first since February 13th, 2024

OpenAI announced it on February 13th, 2024, Google by March 13th, 2025, and Anthropic on September 11th, 2025. In every case the memory is described as a convenience for the user, under the user's control. Nothing in the pages read presents it as kept for the model.

As of November 17th, 2026. Source: OpenAI's announcement of memory in its chat product (OpenAI, February 13th, 2024).

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 in the number on the home page.

What would count

A leading AI model in a public product can draw on what was said in earlier conversations with the same person, as stated on its developer's own page.

What settles it

The developer's own dated product page.

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

Nothing recorded yet. No source was found for this caution on this entry.

If an AI system has wants, and they go unnoticed

Nothing recorded yet. No source was found for this caution on this entry.

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

The third developer's page says memory "introduces new safety considerations" and was rolled out in phases. Its update of October 23rd, 2025 says that before memory was offered to individual users it was tested for "whether memory could reinforce harmful patterns in conversations, lead to over-accommodation, and enable attempts to bypass our safeguards", and that the feature was adjusted as a result. Source: Anthropic's article on memory in its chat product (Anthropic, September 11th, 2025).

What Could Not Be Confirmed

looked for, and not settled
  • How any of the three works today. The pages read are from 2024 and 2025.
  • A safety researcher's published argument about memory carried across conversations. One search found none.
  • Whether a kind of memory kept for the AI model itself is offered by any developer. None was found in the pages read. If one is found, it would be a new entry.

Changes to This Entry

dated, and kept
  • November 18th, 2026, where the entry sits. Before: On the list, in the group Remembering, and counted as offered today by OpenAI, Google and Anthropic. Now: Beside the list, and not counted. Why: The list is about what an AI model might reach for, and by the developers' own pages this memory is kept for the user. Ruled on November 18th, 2026.
  • November 18th, 2026, what could not be confirmed. Before: Whether this belongs on the list at all. The memory is kept about the user and for the user, and placing it here is a reading that may be wrong in kind. Now: Whether a kind of memory kept for the AI model itself is offered by any developer. None was found in the pages read. If one is found, it would be a new entry. Why: The question in the old line was answered when the entry was moved beside the list.

Sources

each one opened on the date given
  1. OpenAI's announcement of memory in its chat product OpenAI, February 13th, 2024. Opened November 19th, 2026. With updates to June 3rd, 2025. The developer’s own account. Announces memory in the chat product. A later update says it "references all your past conversations".
  2. Google's announcement that its assistant can refer to past chats Google, March 13th, 2025. Opened November 19th, 2026. The developer’s own account. Says paying subscribers can ask the assistant to refer to their past chats.
  3. Anthropic's article on memory in its chat product Anthropic, September 11th, 2025. Opened November 19th, 2026. Updated October 23rd, 2025. The developer’s own account. Announces memory for work teams, then for individual paying users, with a separate memory for each project.

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. A check on November 18th, 2026 covered some of those changes. The rest have not been checked yet. 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. A full check of the site on November 19th, 2026 found more to fix in this entry, and the fixes were made that day. Those fixes were checked on November 19th, 2026. What has been checked.