Is Claude private? Training, retention and incognito chats

Claude chats are private in the sense that they are not public and, by Anthropic's own help pages, its staff cannot open them by default. They are not private in the sense of being out of Anthropic's reach: your chats sit on its servers, they can be used to train future models if the model training setting is on, and some are kept for years if they are flagged or rated.

Everything below comes from Anthropic's privacy policy (effective September 10, 2026) and its privacy and support center, checked on November 11, 2026. It covers the consumer plans: Free, Pro, Max and Ultra.

Does Claude train on your chats?

It can. Anthropic's help article Is my data used for model training? lists three cases where your chats and coding sessions may be used to improve its models:

You allow it through the model training setting.
A conversation is flagged for safety review. Anthropic says it may then use it to improve how it detects and enforces its Usage Policy, including training models for its Safeguards team.
You opt in some other explicit way, such as joining its Trusted Tester Program.

A fourth route is easy to miss: the thumbs up and thumbs down buttons. Rating a response stores the entire related conversation for up to five years, and Anthropic says it may use that feedback to train its models.

What counts as "the chat" is broad. Anthropic says it includes the entire conversation, any content in it, custom styles and conversation preferences, and data from Claude for Chrome. It does not include the raw content of connectors such as Google Drive, unless you paste that content into the chat.

The privacy policy lists "Inputs and Outputs from our Services (unless users opt out)" among its training sources, so do not assume the setting is off. Check it.

How to turn off model training in Claude

Anthropic's settings article gives the same path on a computer and on your phone:

Open Claude and select your name to open the settings menu.
Select Settings.
Select Privacy.
Under Help Improve our AI models, switch the toggle off.

What switching it off does: Anthropic will not use your new chats and coding sessions for future training, and it stops using your previously stored chats in future training runs.

What it does not undo: in Anthropic's words, "Your data will still be included in model training that has already started and in models that have already been trained." And if its safety classifiers flag a conversation, it may still be used to improve internal trust and safety models.

If you leave training on, Anthropic says it de-links your chats from your user ID, such as your email address, before they are used, and runs tools to filter or obscure sensitive data, per its page on personal data in model training.

How long does Claude keep your chats?

Anthropic publishes several different clocks. Here they are together, from its retention article:

Situation	How long Anthropic says it keeps the data
You delete a chat	Gone from your history at once, deleted from back-end storage within 30 days
Incognito chat	Not in your history, kept for 30 days
Model training setting on	Up to 5 years in a de-identified form, for new or resumed chats after you switched it on
You rated a response (thumbs up or down)	5 years for the data in that feedback
Chat flagged as violating the Usage Policy	Inputs and outputs up to 2 years, trust and safety classification scores up to 7 years

Chats you never delete stay in your history until you delete them or your account. On top of all of these, Anthropic says it may keep chats "as required by law, to resolve disputes, or as necessary to combat violations of our Usage Policy."

Can Anthropic employees read your Claude chats?

Not by default, according to Anthropic's page on how it protects user data. It says employees cannot access your conversations unless you share them as feedback, or a review is needed to enforce the Usage Policy. In that case, only designated members of its Trust and Safety team can see the data, on a need-to-know basis.

The same page says your data is encrypted in transit and at rest. That is encryption Anthropic controls, which is what lets its systems run Claude on your messages and its reviewers see flagged ones. Anthropic does not describe Claude chats as end-to-end encrypted.

What incognito chats do and do not do

Incognito is the closest thing Claude has to an off-the-record mode. Per Anthropic's incognito article:

How to start one: open a new chat outside of a project and click the ghost icon in the upper right corner. You will see a black border and an "Incognito chat" label. Click the "x" to close it.
Not saved to your chat history or to Claude's memory, and Claude will not pull from it when it searches your past chats.
Not used for training, even if the model training setting is on.
Still kept for 30 days on Anthropic's side.
Still personalized: Claude can see your profile information, such as custom styles and personal preferences, inside an incognito chat.
One way only: you cannot turn an incognito chat into a regular one, and once closed it cannot be reopened. Copy anything you need first.
Not in projects: the ghost icon does not appear in project chats.

On Team and Enterprise plans, incognito chats are included in the organization's data exports for account owners and, on Enterprise, in the Compliance API. Incognito hides a chat from your history, not from your employer.

Memory and chat search

Claude can also remember things across conversations. Anthropic's memory article says memory is on by default for Free, Pro, Max and Ultra plans. To manage it, go to Settings, then Your Claude (or Settings, then Memory in the older layout):

Turn off Generate memory from chats. You then choose Pause memory, which keeps what is saved but stops using it or adding to it, or Reset memory, which permanently deletes all memories, project memories included.
Turn off Search and reference chats to stop Claude searching your past conversations. That search is a paid plan feature.
For a single chat, open the + menu and turn off Memory before you send your first message. Unlike incognito, that chat still stays in your history.

How to delete Claude conversations

From Anthropic's delete or rename article:

On the web: find the chat in the left sidebar or under Chats and tasks, hover, click the ⋮ button, choose Delete, then confirm. To delete several, click Chats and tasks, use ⋮ then Select, tick the chats (or Select all), and click Delete twice.
On iPhone: touch and hold the chat in your list, tap Delete, then confirm.
On Android: in an open chat, tap the ⋮ menu, then Delete, then confirm. For several, tap the checklist icon in your chat list, tick them, and tap the trash icon.

Deleted chats leave your history at once and are erased from the back end within 30 days. Anthropic also says a deleted chat will not be used to train future models. Projects and account information stay until you delete them separately.

Two related clean-ups:

Shared links. A shared chat can be viewed by anyone with the link. Review them under Settings, then Privacy, then Manage next to Shared chats, and click Unshare, per the sharing article.
Your whole account. Download a copy first under Settings, then Privacy, then Export data (web or desktop only, and the emailed link expires after 24 hours). Then go to Settings, then Account, then Delete Account. On a paid plan, Anthropic says to cancel and wait for the billing period to end first. Deletion is permanent, per its account deletion article.

Work accounts and the API are different

The rules above are for personal plans. For its commercial products, including Claude for Work and the Anthropic API, Anthropic says it does not use inputs or outputs to train its models by default, unless you send feedback or otherwise choose to allow it. Your organization's owners, not you, control retention and exports on those plans.

Where your email fits in

Your Claude account, its receipts and its security codes all run through your email address, and your inbox raises the same questions. We wrote about the same questions for Google in does Google AI read my Gmail and for Apple in is Apple Intelligence private.

Secria Mail has no ads and Secria never mines your mail. Your mailbox is locked with a key made from your password, on your device, and sign-up needs no phone number. The free plan is a name, the address you want and a password, here.

Frequently asked questions

Is Claude private and confidential?

Claude chats are private to your account unless you share a link, and Anthropic says its employees cannot access them by default. They are stored on Anthropic's servers under encryption it controls, can be reviewed if flagged for a Usage Policy issue, and can be used for training if the model training setting is on.

Does Claude train on your data if you opt out?

Not your new chats, and not your older stored chats in future training runs. Anthropic says data already in a training run that has started, or in models already trained, stays there. Chats flagged by its safety systems and anything you send as thumbs up or down feedback can still be used.

Does Claude save chats forever?

No fixed forever. Chats stay in your history until you delete them or your account, then leave the back end within 30 days. Longer periods apply if training is on (up to five years, de-identified), if you rated a response (five years), or if a chat was flagged (up to two years, seven for classifier scores).

Is Claude incognito mode really private?

It keeps the chat out of your history, out of memory and out of training. Anthropic still keeps incognito chats for 30 days, Claude can still see your profile preferences in them, and on Team and Enterprise plans they appear in the organization's data exports.

Can Anthropic read my Claude chats?

Anthropic says employees cannot access conversations by default. The exceptions are chats you send as feedback and chats that need review to enforce its Usage Policy, which only designated Trust and Safety staff can see on a need-to-know basis.