Anthropic privacy policy changes: all 13 versions, dated

The Morse teamPublished November 5, 202611 min read

Anthropic's privacy policy changes add up to 13 dated versions of its consumer privacy policy, from July 8, 2023 to the one in force today, effective September 10, 2026. For your Claude chats, three changed the most: March 4, 2024, which promised not to train on your chats; September 28, 2025, which reversed that unless you opt out; and July 8, 2026, which added ID checks, connected apps, and wider grounds for sharing data with law enforcement. The consumer terms haven't changed since October 8, 2025, and a new Usage Policy takes effect on November 12, 2026.

On November 5, 2026, we read every version Anthropic keeps in its own legal archive: the privacy policy, the consumer terms, and the Usage Policy. We compared each version with the one before it, line by line, then checked each against Anthropic's change log and its announcements. Where Anthropic published no summary, we say so, and where its summary leaves something out, we quote the policy.

Anthropic still hosts every version, back to 2023, at its own address on anthropic.com, so we didn't need the Wayback Machine at all. OpenAI's own archive is missing two versions, as our OpenAI privacy policy history found.

When did Anthropic change its privacy policy?

Thirteen times. The table lists every effective date in Anthropic's archive, what the new text changed for a Claude user's chats, and whether Anthropic summarized it.

The 2025 training switch, the five-year retention, and the setting that controls both are covered in depth in is Claude private.

How has Anthropic's promise about training changed?

It went from silence in 2023 to a promise in 2024 to an opt-out in 2025. The July 2023 policy mentioned training on your words only for Feedback, which it said Anthropic would "use under your authorization to train and improve our Services." The first consumer terms in Anthropic's archive, effective February 3, 2024, went further: "We will not train our models on any Materials that are not publicly available," except for feedback and chats flagged for safety review.

The March 4, 2024 privacy policy put the same promise in its own words: "We will not use your Inputs or Outputs to train our models, unless" a chat was flagged for Trust and Safety review, you reported it, or you opted in. That sentence stayed in place through May 2025.

A smaller edit came first. In the December 17, 2024 version, "When you submit Feedback and provide us permission, we disassociate Inputs and Outputs from your user ID" lost the words "and provide us permission," so the policy no longer mentions any permission beyond the rating itself. Then the September 28, 2025 version replaced the promise with "We may use your Inputs and Outputs to train our models and improve our Services, unless you opt out through your account settings." The consumer terms followed on October 8, 2025.

The July 8, 2026 version widened one exception. With training off, chats flagged "for potentially violating our Usage Policy" were still used to train Anthropic's safety classifiers. Now it's chats flagged "for harmful content or for potentially violating our policies."

What changed in Anthropic's privacy policy in 2026?

Three versions took effect. The January 12 version was small: it linked a Consumer Health Data Privacy Policy for users in Washington and states with similar laws who "integrate third party health applications with Claude," and gathered the country sections under one heading.

The July 8 version, announced June 8, was the largest rewrite since September 2025. Anthropic's change log summarizes four parts of it: connected apps and multi-step tasks, Verification Data, study participation, and "more detail about how we keep in touch with you." If Anthropic asks you to verify your age or identity, the policy says it may collect your ID, a photo or video of you, and "facial geometry templates," which our guide to AI age verification compares with eight other apps.

Two more changes in the July text get only a phrase in the summary, which says Anthropic "added clarity around when data may be shared with third parties":

• Law enforcement. The January 2026 text said Anthropic "may disclose personal data to governmental regulatory authorities as required by law." The July text says it "may share personal data with government authorities, law enforcement, or other third parties" where it has "a good-faith belief that disclosure is reasonably necessary," including to "prevent serious harm to any person or to property" or to "enforce our terms." How courts have reached AI chats so far is covered in can AI chats be subpoenaed.
• Work email. A new line says that if you sign up with "an email address owned by your employer or another organization, your account may be linked to the organization's enterprise account." The consumer terms have said this since February 2024, adding that the organization's administrator "may be able to monitor and control the Account, including having access to Materials," but the privacy policy hadn't mentioned it before. What a work account exposes is covered in can my employer see my ChatGPT.

The July version also says, for the first time, that Anthropic has "measures in place to detect and remove children from our Services."

What did the September 10, 2026 update change?

Only the section for users in South Korea. We compared the September 10 text with July 8's line by line, and every other line is the same. As of November 5, 2026, Anthropic's change log has no entry for this version.

The new Korea section lists data Anthropic collects for "Identity verification, Fraud and abuse monitoring": "Name, birthdate, sex, native-foreigner-code, and DI (duplication information)," kept "until account deletion." It names Korea Credit Bureau for identity verification and says the same data goes to Persona in the United States, which keeps it for "3 years." It also replaces Bae, Kim & Lee with Kim & Chang as the firm assisting Anthropic's local representative. If you don't live in South Korea, nothing in your policy changed on September 10.

Does Claude show ads or sell your data?

Anthropic says no to both, and the sale line hasn't changed since 2023: "Anthropic does not 'sell' your personal data as that term is defined by applicable laws and regulations." Its July 2026 summary adds that "Claude remains ad-free."

The sentence about targeted advertising has changed twice, and its heading once. None of it appears in a summary:

1. July 2023: "Anthropic also does not engage in targeted advertising using your personal data."
2. March 2024: "If we share your personal data for targeted advertising to promote our products and services in the future, you can opt-out and we will honor global privacy controls."
3. December 2024: "You can opt-out of sharing your personal data for targeted advertising to promote our products and services," with "in the future" gone. The December change log entry doesn't mention it.
4. July 2026: the same sentence, under a new heading, "Sale & targeted Anthropic marketing of its products and services."

The sentence covers advertising that promotes Anthropic's own products and services. Ads inside Claude are a separate matter, and the July 2026 summary rules them out. To use the opt-out, you can turn on Global Privacy Control, a browser setting that sends the request for you, which the policy says Anthropic honors.

How have Claude's consumer terms and Usage Policy changed?

Less often than the privacy policy. Anthropic's archive has eight versions of the consumer terms, the latest effective October 8, 2025, and four dated versions of the Usage Policy, with the newest taking effect November 12, 2026. Anthropic's privacy change log covers neither, so we list the summary for each where we found one.

The November 12 Usage Policy is the change most Claude users will meet next. Anthropic's announcement, published October 8, 2026, says the abuse rule "does not apply to common versions of user frustration, pushback, dark creative themes, or model testing and research," and ties it to Claude being able to "end rare conversations with persistently abusive users on Claude.ai and Claude Code." The new text also says that "encountering a block does not by itself indicate a violation of the Usage Policy."

What does Anthropic's change log leave out?

Five things, as of November 5, 2026. The change log was last updated June 8, 2026, and it has these gaps:

1. No entry for September 10, 2026. The version in force today has no summary.
2. A list of old versions that stops in 2025. Its "Previous Versions" list ends at September 28, 2025. The October 8, 2025, January 12, 2026, and July 8, 2026 versions are missing from it, though each still loads at its own address and in the version history on the policy page.
3. No summaries before December 2024. The five versions from July 2023 to June 2024 have none. A separate help article covers the spring 2024 updates, including "We've changed our default data retention period to 30 days," but doesn't give dates.
4. Summaries that skip things. The October 8, 2025 entry lists "two minor updates" and leaves out the new South Korea section. The December 2024 entry calls its changes to the legal bases section "non-substantive" and leaves out the advertising line.
5. Two texts with the same date. Two different policies are dated June 5, 2024. The one the change log links adds a Brazil section and says data is protected with "encryption in transit and at rest"; the other is shorter, and it's the one the policy page's version history leads to.

To check a version yourself, open the current policy, use "See version history" under the date, and compare the two texts. Our comparison of 13 apps' change clauses covers what Anthropic promises to tell you before a change.

How Morse keeps your chats out of the next policy update

Each version in these tables set new rules for chats that already sat on Anthropic's servers under your account, and the terms say that if you keep using Claude after an update, "you agree to the updated Terms." Anthropic did limit its 2025 training change to new or resumed chats. Still, the only way to refuse a new version is to stop using Claude. Morse is built so there's no stored history for a new policy to reach. Your chat history is kept in your browser, encrypted, and our servers don't keep your messages or log what you write.

Morse doesn't train AI on your content, so there's no promise to reverse and no opt-out to find. It doesn't sell your data or show ads, and runs no advertising trackers or third-party analytics. Because we don't store chats, we usually have very little to give authorities, whatever grounds a policy lists. Each chat has its own privacy mode:

• Anonymous, to keep using Claude with no Anthropic account behind your words. Our proxy removes your account, email, and IP address and sends the message under Morse's own account. There's no Anthropic account of yours to link to an employer or to hold your history. Anthropic sees the message without knowing it's you, and may keep it for a limited time to check for abuse, under its own policy, so leave names out of anything sensitive.
• Private, for chats you'd rather no model host keep. It's the default for a new chat, 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.
• 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 opening it takes two keys, one of which never leaves your device. The account holds sign-in details, your plan, your balance, and a usage ledger with no content (Morse privacy policy). Clearing your browser deletes your chats for good, and Morse can't recover them, so keep a copy of anything you need.

Try a chat from the Morse home page without an account, or start free with 15 messages a day, Claude Haiku 5.5 included.