Claude privacy: does Claude train on your data, what Anthropic keeps, and where the settings are

If you paste client or company material into Claude at work, three questions matter more than the rest: does Anthropic train on it, how long is it kept, and where is the setting. This page answers all three from Anthropic's own published pages, each quoted with its date, and gives no verdict on whether Claude is private enough for your work. That part is yours, and your firm's.

Short answer: on Free, Pro and Max, whether Claude trains on your data is a setting you control, and the setting also decides how long Anthropic keeps your chats: up to five years if you allow training, 30 days if you don't. On Team, Enterprise and the API, training is off by default; API inputs and outputs are deleted within 30 days, and Team and Enterprise chats stay in the product until you delete them, then leave the back end within 30 days. Everything below is what Anthropic publishes, read on 27 October 2026 and re-read on 6 November 2026, with the article dates Anthropic shows; the sources are listed at the end. This page is about training, retention, the privacy settings and the policy behind them. For who can open a chat (sharing, your organization's owners, Anthropic's staff), read is Claude confidential?, which this page does not repeat.

Does Claude train on your data? Anthropic's answer, by plan

Anthropic answers this twice, in two Privacy Center articles with the same title: one for what it calls consumer products, "such as Claude Free, Pro, Max", and one for commercial products, meaning Claude for Work (the Team and Enterprise plans) and the Anthropic API. Which article applies to you depends on the account you are signed in to, not on which window you type into. As of 6 November 2026:

Two details from the consumer article matter at work. First, what counts as training data when the toggle is on is the whole exchange: Anthropic says it stores "the entire related conversation", and the article names custom styles and preferences as part of it. Second, what is left out: "It does not include raw content from connectors (e.g. Google Drive)", unless that content is copied into the chat itself. A pasted paragraph is in; a document Claude fetched through a connector is not, though Claude's answer about it is. The article also states a rule that holds regardless of the toggle: "Your Incognito chats are not used to improve Claude, even if you have enabled Model Improvement in your Privacy Settings."

Anthropic's current consumer Privacy Policy (effective 10 September 2026) states the same position in legal form: "We may use your Inputs and Outputs to train and improve Anthropic AI models" "unless you opt out through your account settings", followed by "Even if you opt-out, we will use Inputs and Outputs for model improvement when:" and a list that covers safety-review flags and reports you make to Anthropic directly.

What changed in 2025, and what Anthropic's pages say today

Before 28 August 2025, a consumer Claude chat ran on what Anthropic's post of that day calls its "existing 30-day data retention period"; TechCrunch reported that previously Anthropic "didn't use consumer chat data for model training". On that day Anthropic published "Updates to Consumer Terms and Privacy Policy", introducing a choice for Free, Pro and Max users, Claude Code on those accounts included, and leaving commercial services out: the post says the updates "do not apply to services under our Commercial Terms", naming Claude for Work, Claude for Government, Claude for Education and API use.

Three things in that post still shape the terms today. The retention change: "We are also extending data retention to five years, if you allow us to use your data for model training", while "If you do not choose to provide your data for model training, you'll continue with our existing 30-day data retention period". The scope: "These updates will apply only to new or resumed chats and coding sessions", so "Previous chats with no additional activity will not be used for model training". And the control: "You can change your selection at any time in your Privacy Settings."

How the choice was presented is a matter of record. TechCrunch reported on 28 August 2025 that the new policies apply to "Claude Free, Pro, and Max users, including those using Claude Code", that "Business customers using Claude Gov, Claude for Work, Claude for Education, or API access will be unaffected", and that existing users saw a pop-up headed "Updates to Consumer Terms and Policies" with a large Accept button and, below it, a training toggle "automatically set to 'On'". TechCrunch gave the decision date as 28 September 2025; Anthropic's post, as it reads today, gives existing users "until October 8, 2025 to accept the updated Consumer Terms and make your decision". If you accepted that pop-up quickly in autumn 2025, the toggle may still be on; the next section is how to look.

Today's retention terms, as Anthropic's pages state them on 6 November 2026:

Zero data retention is a phrase people meet when they look into Claude privacy, so here is what Anthropic says it is and is not. Its Privacy Center article describes arrangements under which "Anthropic does not store their inputs or outputs except where needed to comply with law or combat misuse or harm", available to "Some Claude Platform (Claude API) and Claude Code for Enterprise customers, subject to Anthropic's approval", and states that "the only products to which zero data retention applies are eligible Anthropic APIs, Anthropic products that use your Commercial organization API key (including Claude Code accessed via the API), and Claude Code for Enterprise plans" (read 6 November 2026). The Claude Code documentation adds: "ZDR is not included in the standard Enterprise plan; it is enabled on a per-organization basis by your account team after confirming eligibility." Nothing on those pages offers it to Free, Pro or Max.

How to check and change the training setting

These are Anthropic's steps, from its Privacy Center article "How do I change my model improvement privacy settings?" (dated 3 August 2026), with its labels as written:

1. "Select your name from your settings menu and click the drop down button" (on the phone apps, select your name without the drop-down).
2. "Select Settings".
3. "Select Privacy".
4. "Under Help Improve our AI models, click the button to toggle it off/on".

Anthropic's Claude Code documentation gives the same screen a direct address, claude.ai/settings/data-privacy-controls, which is the quickest check from a laptop. If you are on a Team or Enterprise account, you will not find the toggle: the commercial default applies and the choice is your organization's, not yours. Older Anthropic pages call the same control "Model Improvement", and the consumer training article still uses that name.

What switching it off does, in Anthropic's words: "we will not use any new chats and coding sessions you have with Claude for future model training", and previously stored chats are left out of future runs. What it does not do: "Your data will still be included in model training that has already started and in models that have already been trained." Chats flagged by safety classifiers may still be used for trust and safety work. The toggle is per account, not per chat, so it covers every conversation and coding session on that account at once, and it covers nothing on a second account you are also signed into.

Claude privacy settings and Incognito: what the labels do

The training toggle is one of a handful of controls, and the others sit on the same Privacy screen or one tap from a new chat. Anthropic's Privacy Center groups them under a collection it calls "Privacy Settings & Controls"; its Help Center covers incognito chats separately. Here is each label as Anthropic's pages show it on 6 November 2026, and what Anthropic says it does. None of this is a judgement on whether the controls are enough for your work.

Incognito is the control people at work ask about most after the training toggle, so here is the rest of what Anthropic's Help Center article says, because three of its lines change what the mode is good for. First, it is one-way: "once you start an incognito chat, it cannot be converted to a regular chat or saved to your history", and "Once closed, incognito chats cannot be reopened." Second, it is not available inside a project: "Incognito mode is currently only available for chats outside of projects, so you won't see the ghost icon when starting a chat within a project." Third, on a work account it is not invisible to your firm: "Incognito chats are included in organizational data exports available to account Owners", and "Incognito chats are included in the Compliance API (available for Enterprise plans)". The article also carries a note at the top as read on 6 November 2026: "If you have the new Claude experience, incognito chats open in the previous chat experience, so Claude can't create files or run code in them." Your profile information still applies inside one: asked whether Claude can access custom styles and personal preferences in an incognito chat, Anthropic answers "Yes".

What the privacy policy says, in its own words

The Privacy Center articles above are Anthropic's explanations; the document behind them is its Privacy Policy at anthropic.com/legal/privacy, which on 6 November 2026 carries the line "Effective Sep 10, 2026" and is marked for the audience "Individual". The Privacy Center's "How do you use personal data in model training?" article adds the boundary: "the Privacy Policy does not apply where Anthropic acts as a data processor and processes personal data on behalf of commercial customers", in which case "the commercial customer is the controller". So the policy is the consumer document; a Team or Enterprise seat sits under your organization's agreement with Anthropic. The sentences that bear on this page, quoted from the policy as read that day:

• Training. "We may use your Inputs and Outputs to train and improve Anthropic AI models, unless you opt out through your account settings. Even if you opt-out, we will use Inputs and Outputs for model improvement when: (i) your conversations are flagged for safety review to improve our ability to detect harmful content, enforce our policies, or advance AI safety research, or (ii) you've explicitly reported the materials to us (for example via our feedback mechanisms)."
• What "Inputs and Outputs" covers. The policy's definition names "chat, coding, agentic sessions (where Claude performs multi-step tasks or takes actions on your behalf), and connected services", and the content "you choose to upload".
• Retention. "Anthropic retains your personal data for as long as reasonably necessary for the purposes and criteria outlined in this Privacy Policy and explained further in our privacy center." The specific periods are the ones in the table above, from the Privacy Center.
• Deleting a chat. Under "Rights and Choices": "You also are able to delete individual conversations, which will be removed immediately from your conversation history and automatically deleted from our back-end within 30 days."
• Feedback and flagged chats. "When you submit Feedback, we disassociate Inputs and Outputs from your user ID to use them for training and improving our models." For flagged content: "we disassociate the content from your user ID to train our trust and safety internal classification and generative models. However, we may re-identify the Inputs or Outputs to enforce our Terms of Service or Usage Policy with the responsible user if necessary."
• Selling data. "Anthropic does not 'sell' your personal data as that term is defined by applicable laws and regulations."

Anthropic keeps a running summary of what each version changed, in its Privacy Center article "Updates to our Privacy Policy". As read on 6 November 2026 the most recent summary there is headed "effective July 8, 2026" and says those updates "apply only to consumer accounts (Claude Free, Pro, and Max plans)"; it lists more detail on "Multi-step tasks and connected apps", on "Verification Data", on "Study participation", and closes: "While our products have evolved, our commitments haven't: We don't sell your data, Claude remains ad-free, and you can control whether your conversations are used to improve Anthropic's AI models." The summary that introduced the training choice is headed "effective September 28th, 2025": "we may use your chats and coding sessions to improve Claude, if you choose to allow us to", and "If you choose to allow us to use your data to improve Claude, we'll retain this data for 5 years." The summary page lists no entry for the 10 September 2026 version the policy itself is dated with, so what that version changed is not something this page can say; the policy's own text is what is quoted above.

Claude Code: a different reader, sent to Anthropic's documentation

Claude Code runs on the same accounts as the chat app, so people who look into Claude privacy soon meet its documentation, and a common worry is that code written on a Pro or Max plan is treated differently from what people assume. Anthropic's own data-usage page settles the point in one sentence: it will train "using data from Free, Pro, Max, and Ultra accounts when this setting is on (including when you use Claude Code from these accounts)" (read 6 November 2026). Under commercial terms it is the opposite default: Anthropic "does not train generative models using code or prompts sent to Claude Code under commercial terms, unless the customer has chosen to provide their data to us for model improvement". Retention follows the same two lines as chats (5 years if allowed, 30 days if not), and the page adds that Claude Code keeps session transcripts on your own machine for 30 days by default. If your question is about a development team, telemetry, cloud sessions or a Bedrock or Vertex deployment, that page at code.claude.com is the right reader; this guide stays with the person pasting a client's email into a chat.

What people at work choose not to paste

None of this is a verdict on Claude, and none of it is advice to put client material into any tool your firm or your engagement terms rule out. These are the practices people in client work describe once they have read the terms above:

• Check the account before the paste, not after. The five-year term starts with the chat. A quick look at Settings → Privacy, or at which account the browser is signed into, costs ten seconds; the choice cannot be made retroactively for a chat already used in a training run.
• Treat a thumbs-up on a client chat as sending that chat to Anthropic. By Anthropic's pages, rating a response stores "the entire related conversation" for up to five years on every plan, whatever the toggle says. People who rate answers do it on chats with nothing in them they would mind keeping.
• Describe the matter, not the client. "A fixed-fee engagement where the client wants scope added without a change order" carries the question without the name, the figures or the thread. The retention term applies to whatever you typed, so the less identifying the input, the less there is to keep.
• Let the firm's account carry the firm's work. A Team or Enterprise seat runs under the commercial default; a personal Pro account runs under a setting that one Accept click in 2025 may have left on, as TechCrunch described the pop-up. Mixing them means two sets of terms on one matter.
• Material under someone else's terms is their call. An NDA, a data-room agreement or an engagement letter that limits third-party processing decides before any setting does.

If your question is about what Claude carries from one conversation to the next, that is a separate mechanism from training; see does Claude remember previous conversations?

Where Equerry fits: what it reads, and what you can see and delete

Equerry is a personal assistant who helps you get things done, and it reads only what you decide to send it: an email thread you CC or forward to your own Equerry address, a PDF, image, link or text you share from your iPhone's share sheet, a note you type, or a voice note you record in Equerry. With your permission it also reads your calendar, and only reads it. It never asks for your inbox. Our privacy page states the training position in one sentence, "We never train AI models on what you save", and our AI disclosure names the outside providers that process what you send and what each receives. Here is what that looks like for the question this page is about:

The point for a trust-at-work reader is the shape of the decision, not a safety claim. With Equerry there is no account-wide toggle to find, because there is no training to switch off and no inbox connection to scope: each item is in because you sent it, and each extracted fact is visible and deletable by you. What stays with our providers is set out on the privacy page as well: forwarded emails are held by our email provider for 30 days, and when you delete your account your data leaves our live database straight away and our backups within 40 days. Equerry sends nothing, books nothing and buys nothing on your behalf, and makes no claim about whether anything you send it is confidential. Which tool, if any, a client's material belongs in is a call you make against your firm's rules and the terms that client set, and this page does not make it for you.

Sources, read 2026-10-27 and re-read 2026-11-06. Anthropic: Updates to Consumer Terms and Privacy Policy (28 August 2025); Privacy Policy (effective 10 September 2026). Anthropic Privacy Center: Is my data used for model training? (consumer, dated 16 March 2026); Is my data used for model training? (commercial, dated 18 August 2026); How long do you store my data? (dated 1 July 2026); How long do you store my organization's data? (dated 1 July 2026); How do I change my model improvement privacy settings? (dated 3 August 2026); Can you delete data sent via Claude? (dated 16 March 2026); Export your Claude data (no fixed date shown); How do you use personal data in model training? (dated 16 March 2026); Updates to our Privacy Policy (dated 8 June 2026); I have a zero data retention agreement with Anthropic. What products does it apply to? (no fixed date shown). Anthropic Help Center: Use incognito chats (no fixed date shown). Claude Code documentation: Data usage (no date shown). Press, for the 2025 history only: TechCrunch, Anthropic users face a new choice: opt out or share your chats for AI training (28 August 2025). Each Anthropic article names the plans it covers (Free, Pro and Max on the consumer side; Team, Enterprise and the API on the commercial side); if Anthropic's current page and this one differ, follow Anthropic's page. Equerry statements come from our own privacy and AI disclosure pages. This guide is general information and is not legal advice.

Last updated 2026-11-06.

All product names, logos, and brands are property of their respective owners. Equerry is not affiliated with, endorsed by, or sponsored by any other product mentioned; comparisons are provided for informational purposes.