Claude AI Privacy: Training, Retention and Deleting Your Data

Claude AI privacy on a personal account (Free, Pro or Max) comes down to one setting and a few retention rules. As of September 2026, Anthropic’s Privacy Policy says it may use your chats to train its models unless you turn off the model improvement setting in your privacy settings, and even then it may use chats flagged for safety review or ones you send as feedback. Deleted chats leave your history immediately and are removed from Anthropic’s back-end within 30 days. Business plans and the API follow different rules: by default, Anthropic doesn’t train on them.

Every fact below comes from Anthropic’s own pages: its Privacy Policy (effective September 10, 2026), its Privacy Center and its Help Center, all checked in September 2026. Where an article has its own date, we give it. Anthropic updates these pages, so check the current versions before relying on a detail.

Does Claude use your chats for training?

On consumer plans, it depends on your setting. Anthropic’s Privacy Center article “Is my data used for model training?” (dated March 16, 2026) says Anthropic will use your chats and coding sessions, including to improve its models, if any of these apply:

• You’ve chosen to allow your chats and coding sessions to be used to improve Claude.
• Your conversations are flagged for safety review. Anthropic says it may then use them to improve its ability to detect and enforce its Usage Policy, including training models used by its Safeguards team.
• You’ve explicitly opted in some other way, such as joining its Trusted Tester Program.

The same article says training data from a chat includes the whole conversation, any content, custom styles and conversation preferences, and data collected when using Claude for Chrome. It says it does not include raw content from connectors such as Google Drive, or from remote and local MCP servers, unless that content is copied directly into the conversation.

These consumer rules also apply when a Free, Pro or Max account is used with Claude Code, according to the same page.

How to turn the model training setting off

Anthropic’s article on changing model improvement settings (dated August 3, 2026) says the setting is in Settings, under Privacy, labeled “Help improve our AI models”. It’s available on the web, desktop and mobile apps. Turning it off has these effects, per that article:

• New chats and coding sessions won’t be used for future model training.
• Previously stored chats won’t be used in future training runs.
• Your data will still be included in training that has already started and in models already trained.
• Conversations flagged by safety classifiers may still be used to improve internal trust and safety models.

You can switch it back on at any time. If you do, Anthropic’s retention article says the longer training retention period described below applies only to new or resumed chats after you enable it.

How long Anthropic keeps your data

Anthropic’s Privacy Center article “How long do you store my data?” (dated July 1, 2026) sets out these consumer retention periods:

Situation	What Anthropic says it keeps
You delete a chat	Removed from your history immediately; deleted from back-end storage within 30 days
Model improvement is on	Data may be kept in a de-identified form for up to five years in model training pipelines
A chat is flagged for violating the Usage Policy	Inputs and outputs kept for up to two years; trust and safety classification scores for up to seven years
You give feedback (thumbs up or down, bug reports)	Data from that submission kept for five years
Legal requirements or disputes	Chats may be kept as required by law, to resolve disputes, or to combat Usage Policy violations

Two details matter here. Anthropic says a chat you delete won’t be used to train future models. And it says it de-links feedback from your user ID before use, and doesn’t combine feedback with your other conversations.

Incognito chats

For a conversation you don’t want kept in your history, Anthropic offers incognito chats. As of September 2026, its Help Center says incognito chats:

• Aren’t saved to your chat history or to Claude’s memory, and aren’t used for training, even if model improvement is on.
• Are retained for 30 days by default, or longer under an Enterprise organization’s custom retention setting.
• Don’t use Claude’s existing memory, although Claude can still see profile preferences such as custom styles.
• Are started from the ghost icon in a new chat outside a project; they can’t be used inside projects.
• Can’t be converted into regular chats or reopened once closed.

On Team and Enterprise plans, the Help Center adds that incognito chats are included in organizational data exports available to account owners.

How to delete Claude conversations

Anthropic’s article on deleting conversations (dated March 16, 2026) describes two routes:

1. One chat. Open the conversation, click its name at the top of the screen, and choose Delete.
2. Several chats. Open your full list of recent chats, tick the ones you want to remove, and choose to delete the selected chats.

Deleted chats disappear from your history right away and are removed from Anthropic’s back-end within 30 days, subject to the exceptions in the retention table above.

Unshare any chats you’ve shared

Anthropic’s Privacy Center article on sharing (dated June 15, 2026) says chats are private by default. If you share one, anyone with the link can see a snapshot of the conversation up to that point, including artifacts, but not attached files or raw data from MCP tool calls. To stop sharing, change the chat’s visibility from Public to Private in its Share menu. On Free, Pro and Max plans, you can review every shared chat under Settings, then Privacy, and unshare each one. On Team and Enterprise plans, chats can only be shared within the organization.

How to export your data and delete your account

Export first. Anthropic’s export article (dated July 8, 2026) says Free, Pro and Max users can export their user data and chat history from Settings, then Privacy, on the web app or Claude Desktop. It isn’t possible from the iOS or Android apps. You’ll receive a download link by email, which expires 24 hours after delivery. On Team and Enterprise plans, only the organization’s Primary Owner can export data.

Then delete the account. Anthropic’s article on deleting Claude accounts (dated March 16, 2026) says:

• You delete your account from Settings, under Account.
• On a paid Pro or Max plan, you first cancel the subscription and wait until the end of the current billing period.
• Deletion is permanent and you lose access to saved chats.
• If you use Claude through a third-party service, you request deletion through that service.

Business plans and the API work differently

Anthropic’s commercial products, which it describes as including Claude for Work (Team and Enterprise), the Anthropic API and Claude Gov, have separate terms. As of September 2026, its Privacy Center says:

• Training. By default, Anthropic won’t use inputs or outputs from commercial products to train its models, unless you report feedback or otherwise choose to allow it (article dated August 18, 2026).
• API retention. Anthropic automatically deletes API inputs and outputs from its back-end within 30 days, except where you use a feature with longer retention under your control, have a separate agreement such as zero data retention, it needs them longer to enforce its Usage Policy, or the law requires (article dated July 1, 2026).
• Team and Enterprise chats. Chats are kept in the product so you can continue them; deleted chats leave your history immediately and the back-end within 30 days.
• Admin controls. Owners of Team and Enterprise plans can disable the feedback buttons for their members.

The practical point: if your employer provides Claude through a business plan, your chats are governed by your employer’s agreement and admin settings, not the consumer setting. If you use a personal account for work, the consumer rules apply, not your employer’s business agreement.

A worked example

This is an illustrative scenario, not a real client. A small-business owner has used a personal Claude Pro account for a year, with model improvement on, to draft contracts that include customers’ names and addresses.

1. She exports her data from Settings on the web so she keeps her useful templates.
2. She turns off “Help improve our AI models”. Her stored chats won’t be used in future training runs, though anything already in a training run or trained model stays there.
3. She deletes the chats that contain customer details, which removes them from her history now and from Anthropic’s back-end within 30 days.
4. She checks her shared chats list and unshares an old contract draft she’d sent to her accountant.
5. For future drafts, she uses placeholders such as “[Customer]” and considers a business plan if her staff start using Claude too.

What she can’t undo is any use in training that happened while the setting was on. The lasting fix is the new habit.

What Claude says about you versus what it stores

Your privacy settings govern your own chats. They don’t control what Claude says about you when someone else asks. Anthropic’s Privacy Center article on training data rights (dated July 8, 2026) says its models are trained partly on public web data, which can incidentally include names and biographical details. It says people covered by laws such as the GDPR and UK GDPR may have rights to object, or to request erasure or correction, but that there are limits on its ability to identify a specific person’s data in training material. If a Claude response includes your personal data, Anthropic asks you to contact it with your name, the prompt used and a screenshot or details of the response.

For US readers, those GDPR rights generally don’t apply; whether a state privacy law gives you similar rights depends on where you live. Our guide to making a data subject access request explains how these requests work, and our guide to your reputation in AI search covers fixing the public sources AI tools draw on. If the problem is persistent, our personal reputation management service can help.

Common mistakes

• Assuming training off means nothing is stored. Chats stay in your history until you delete them.
• Rating a sensitive chat. Feedback stores the whole conversation for five years.
• Forgetting shared links. A shared snapshot stays viewable until you unshare it.
• Using a personal account for client or patient data. The consumer terms apply to it, not a business agreement.
• Deleting your account before exporting. Deletion is permanent.

For how other assistants compare, see our guides to ChatGPT privacy and deleting ChatGPT history.

Frequently asked questions

Is Claude private?

Your chats are private to your account by default and only visible to others if you share them. As of September 2026, Anthropic may use consumer chats for training unless you turn off the model improvement setting, and it retains some data, such as flagged chats and feedback, for set periods described in its Privacy Center.

Does Anthropic train on my Claude chats if I opt out?

Not your new or stored chats in future training runs, according to Anthropic’s pages as of September 2026. Exceptions are chats flagged for safety review and material you send as feedback. Data already in a training run or a trained model stays there.

How long does Claude keep deleted chats?

Anthropic says deleted chats are removed from your history immediately and from its back-end within 30 days, unless they were flagged for a Usage Policy violation, submitted as feedback, or must be kept by law.

Are Claude Team, Enterprise and API data used for training?

By default, no. As of September 2026, Anthropic says it doesn’t train on inputs or outputs from its commercial products unless you report feedback or otherwise choose to allow it.