OpenAI privacy policy changes: every version since 2023, dated

The Morse teamPublished November 4, 20268 min read

OpenAI's privacy policy changes add up to 11 dated versions of its main consumer privacy policy (the US version since 2026) between March 2023 and November 2026, five of them in 2026. The changes that mattered most to ChatGPT users were training on your chats (April 2023), work account administrators seeing your content (2023 and 2024), and, in 2026, ads for Free and Go users and data shared with marketing partners, something the policy had said until April 2026 that OpenAI didn't do.

On November 4, 2026, we read every version we could find. OpenAI still hosts several at dated addresses under its "Previous version" links. For the versions it no longer hosts, we read copies of OpenAI's own policy page saved by the Internet Archive's Wayback Machine, and we used the Wayback Machine for nothing else. We compared each version with the one before it, sentence by sentence, and matched it to OpenAI's own announcements where one exists.

OpenAI keeps more of this history public than most AI apps. It hosts old versions at dated addresses, and for its January 2024 update it published a plain summary in its Help Center.

When did OpenAI change its privacy policy?

Eleven times that we can date, starting from the March 14, 2023 version, the earliest we could read. The table lists each version with the change that mattered most to users, plus four OpenAI announcements that changed what the policy covers.

Two topics have their own pages. Whether ChatGPT trains on your chats, and how to stop it, is covered in does ChatGPT train on your data. The 2025 court order that kept deleted chats ran under the policy's existing legal exception, which the February 2026 version words as keeping data "even after you delete it, for example because we are legally required to"; our court order timeline has the dates.

What changed in OpenAI's privacy policy in 2026?

Ads, and then marketing. In its January 16 announcement, OpenAI promised "We keep your conversations with ChatGPT private from advertisers." The February 9 policy, published the day ad testing began in the US according to OpenAI's ads help page, added ad personalization for Free and Go users and said OpenAI "may receive information from advertisers and other data partners," with the example "we could receive information about purchases you make from these advertisers." The same version added contact uploads: "we upload information from your device address books and check which of your contacts also use our Services."

The April 30 version made the bigger change. Every version from November 2024 on had said "we do not process Personal Data for 'targeted advertising' purposes." The April 30 version deleted that sentence and added that OpenAI "may share limited data with select marketing partners for purposes of promoting our products and services to you on third-party properties. This is known as 'targeted advertising'." OpenAI's help page on that sharing says "We do not share your conversations with these marketing partners," and the policy offers an opt-out in your account settings or through Global Privacy Control. What OpenAI does share, and with whom, is covered in does ChatGPT sell your data.

The September 10 version turned "may receive information from advertisers" into "receive," a change our comparison of 13 apps' policies covers. It also added that for Free and Go users OpenAI collects "your ads history and interests, such as the types of ad content that you view or engage with." The ad settings themselves are in our guide to turning off ChatGPT ads.

Outside the US, EEA, UK, and Switzerland, the rest-of-world policy, dated February 6, 2026, still says on November 4, 2026 that OpenAI does "not process Personal Data for 'targeted advertising' purposes."

Where can you read OpenAI's old privacy policies?

Partly on OpenAI's own site, and the rest only in the Wayback Machine. Click Previous version under the date at the top of the current policy, and you can step back one version at a time. As of November 4, 2026, that path is incomplete in three places:

1. The April 30, 2026 version is skipped. The May 18 version's "Previous version" link goes straight to April 1. The April 30 version still exists at its own address, but nothing links to it, so the version that first added marketing partners doesn't appear in OpenAI's chain. Wayback copies of OpenAI's page from May 1 to May 17, 2026 show "Updated: April 30, 2026" at the top.
2. The April 1, 2026 version links to itself. Its "Previous version" link points to its own address, so the chain stops there. The February 9 version, the first with ads, is reachable only by typing its address.
3. Two versions are missing. The June 27, 2025 version's "Previous version" link jumps to June 2023. We found no page on openai.com with the November 14, 2023 or the November 4, 2024 text, the versions that added administrator access to your content. Wayback copies of OpenAI's page are the only record we found of either.

OpenAI's revision addresses also fill in gaps on their own. An address with a date that has no version of its own still loads a policy: the one for March 1, 2026 shows the February 9 version, so a dated link isn't proof a version was published that day. To check a date, look at the "Updated" line on the page itself. Our guide to downloading your AI chat data covers getting a copy of what OpenAI holds about you, which no policy version lists in full.

Did memory and Temporary Chat change the privacy policy?

Both features came first, and the policy followed months later. Temporary Chat arrived with OpenAI's February 13, 2024 memory announcement, but the policy didn't mention it until the November 4, 2024 version, which said temporary chats are kept up to 30 days. The February 2026 version added that you can "delete specific Saved Memories," and the September 2026 version calls them "Memories."

Both features now affect ads. OpenAI's ads help page says that with ad personalization on, "If memory is on, ChatGPT may save and use memories and reference recent chats when selecting an ad," and that "Temporary Chats will not show ads." OpenAI's Temporary Chat page still says it "may keep a copy of a temporary chat for up to 30 days for safety purposes."

How Morse keeps a policy rewrite away from your chats

Each version in the table sets the rules for data OpenAI already holds: chats, memories, ads history, and contacts, kept on its servers and tied to your account. Morse is built so your chat history isn't on our servers to begin with. It's stored in your browser, encrypted, and our servers don't keep your messages or log what you write. The account holds sign-in details, your plan, your balance, and a usage ledger with no content (privacy policy).

The 2023 training clause has no counterpart here, because Morse doesn't train AI on your content. The 2026 changes have none either. Morse doesn't sell your data or show ads, and runs no advertising trackers or third-party analytics. Temporary chats on Morse are kept in memory only. Because we don't store chats, we usually have very little to give authorities.

The privacy mode you pick for each chat decides who sees the message on its way to a model:

• Private, so no model host keeps a copy for any policy to cover. A new chat starts here, with Grok and open-weight models like DeepSeek, Kimi, and GLM. Morse has zero data retention agreements with its inference, search, and other providers, so they keep nothing. Where a provider confirms zero retention on each reply, Morse checks it and refuses a reply that doesn't.
• Anonymous, to keep using GPT without an OpenAI account behind your words. Our proxy removes your account, email, and IP address and sends the message under Morse's own account. OpenAI sees the message without knowing it's you, and may keep it for a limited time to check for abuse, under its own policy. That policy can change like the ones above, so leave names out.
• E2EE, for anything no one should read. Your browser checks the enclave's hardware attestation, then encrypts the message so only that enclave can open it, and our servers relay ciphertext they can't read. This rests on the enclave hardware and attestation working as designed. E2EE chats come with Pro, at $19 a month.

In every mode, history locks when you sign out and takes two keys to open, one of which never leaves your device. Clearing the browser deletes your chats and Morse can't recover them. Keep a copy of anything you'll need.

You can try a chat from the Morse home page without an account, or start free with 15 messages a day and no card.