Does Claude Store Your Conversations?
Anthropic's Data Retention Explained

Published July 9, 2026 · Updated August 17, 2026 · Facts last verified August 17, 2026

Short answer: yes, Claude can store your conversations — and in 2025 the rules around that changed in a way a lot of people missed. Whether a de-identified copy of them can sit in a training pipeline for up to five years, and how quickly a deletion actually takes effect, now depends on a setting you were asked to choose and on which Claude product you are using.

This guide explains Anthropic's data retention in plain English: what Claude keeps, for how long, what the 2025 policy change actually did, how consumer Claude differs from the business API, and what really happens when you delete a chat. Policies like these change, so treat this as an informed starting point and confirm the current terms for your own account.

The Short Answer: Yes — But the Details Changed in 2025

By default, your normal Claude conversations are saved to your account so you can scroll back and continue them later — with the exception of incognito chats, which are not written to your history at all. That part is unremarkable — it is the same "notebook on your desk" behavior you would expect from any chat app.

What makes Claude worth a closer look is a policy shift Anthropic announced in August 2025. It changed two things at once: whether your consumer chats are used to train Claude, and how long they are kept. Those two questions are now tied to a single choice you make in your settings — so the honest answer to "does Claude store my conversations?" is "yes, and how long depends on you."

The 2025 Policy Change You Should Know About

On August 28, 2025, Anthropic updated its Consumer Terms and Privacy Policy. The headline change: conversations and coding sessions on the consumer plans — Claude Free, Pro, and Max — are now used to help train Anthropic's models unless you turn that setting off.

This was a genuine reversal. Before the change, Anthropic did not train on consumer conversations by default, and that was something people often pointed to as a privacy advantage. Now the default leans the other way, and the control is opt-out rather than opt-in.

A few specifics worth knowing:

• Existing users were asked to accept the updated terms and pick a preference by October 8, 2025. After that date, making a choice became required to keep using Claude.
• New users choose during sign-up.
• The setting applies to new or resumed chats and coding sessions — old conversations you never touch again are not swept in.
• You can change the preference at any time in your Privacy Settings, and it takes effect going forward.

30 Days or 5 Years? Both Numbers Mean Something Narrower Than You Think

These two figures are quoted everywhere, usually as if they were the lifespan of your chat history. Neither is. Getting them right matters, because the common summary — "opt out and your chats vanish after a month" — is not what Anthropic says.

• Your chats stay until you delete them. Opting out of model improvement does not put your history on a 30-day timer. A conversation sits in your account, available to scroll back to, until you remove it.
• The 30 days is a deletion window, not a lifetime. When you delete a conversation it disappears from your history immediately and is removed from Anthropic's back-end systems within about 30 days.
• The five years applies to training pipelines, in de-identified form. If you allow model improvement, Anthropic may keep that material in its training pipelines for up to five years — a different store, holding a de-identified copy, on its own clock.
• Safety-flagged material runs longer still. See the exception below: up to two years for the inputs and outputs, and up to seven years for the trust-and-safety classification scores.

So the training setting is still the most important lever on a consumer plan — it decides whether a de-identified copy of your conversation goes into a five-year pipeline. What it does not do is sweep your visible history away for you. That part is still a button you have to press.

Saved History vs Training Data: Not the Same Thing

As with any chatbot, it helps to separate two ideas that get blurred together.

Saved history is the conversation sitting in your own Claude account for you to revisit. It being saved does not, by itself, mean anyone is studying it.

Training data is conversation content being used to help improve Anthropic's models. On consumer plans this is what the 2025 setting governs. Turn it off and your new chats are, as a rule, no longer used for training — but the rule has two written exceptions, and they are worth knowing rather than discovering. Anthropic's privacy policy effective 8 July 2026 states: "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." So an opt-out is a strong default, not an absolute.

The distinction matters because "Claude saved my chat" and "Claude is training on my chat" are different statements with different answers — and on a consumer plan, you control the second one.

Consumer Claude vs the Business API: What Applies Where

This is the part most summaries skip, and it changes everything. The 2025 consumer change — training unless you opt out, and up to five years in the training pipeline if you allow it — applies to the consumer products. According to Anthropic, it does not apply to:

• Claude for Work (Team and Enterprise)
• The Claude API (including access via Amazon Bedrock and Google Cloud)
• Claude for Government
• Claude for Education

Under these commercial and API terms, Anthropic's stance is markedly stricter: your inputs and outputs are not used to train its models without your explicit permission, and API conversation content is not retained by default. There are named exceptions rather than vague ones: Claude Fable 5 and Claude Mythos 5 are Covered Models that require a 30-day retention window and are not eligible for Zero Data Retention, and features such as the Files API keep what you upload until you delete it. Claude for Work, meanwhile, does store chats so that conversations remain saved and continuable, under the organisation's own retention controls. For organizations that need it, Anthropic also offers Zero Data Retention (ZDR) arrangements, where data is not stored after the response is returned, and HIPAA-ready API access with a signed Business Associate Agreement.

So "how does Claude handle my data?" genuinely has two answers depending on how you reach it: a consumer account and the business API are different worlds.

What Happens When You Delete a Chat or Your Account

Deleting a conversation removes it from your visible history — the immediate effect most people are after. Anthropic also states that a conversation you delete will not be used for future model training. Changing your training preference or deleting your account likewise excludes your data from future training.

One consequence of Claude's memory feature deserves its own line, because it is the same trap ChatGPT users hit. Since July 2026 Claude stores memory as individual categorised entries rather than daily summaries, and Anthropic's documentation is explicit about what deletion does not reach: "When a conversation expires or is deleted, related memory entries generated from it won't be removed, but you can delete individual memories at any time." Deleting the chat does not delete what was learned from it — the memory entry is a separate object you have to remove separately.

As always, be precise rather than over-promising about the back end. Deletion and purge timing can depend on the product, on legal obligations, and on security needs, and providers commonly keep short-lived copies for a limited window before permanent removal. Do not assume "deleted" means "erased from every system the instant you click." For the exact timing that applies to your account, check Anthropic's current documentation.

The Exception: Safety-Flagged Conversations

There is one caveat that applies even when you have opted out, and even under a Zero Data Retention arrangement. If a chat is flagged by Anthropic's automated trust-and-safety systems, Anthropic may retain the inputs and outputs for up to two years, and the trust-and-safety classification scores for up to seven years. As set out above, flagged material may also still be used for model improvement even if you opted out.

This is standard across major providers and is not unique to Claude, but it is worth knowing: the deletion window described above is the ordinary case, not a guarantee that nothing is ever kept longer. It applies even under a Zero Data Retention arrangement.

What About Uploaded Files?

Files you share with Claude — PDFs and images — are handled according to the same retention and training rules as the conversation they belong to. They are processed to answer your request and kept according to the policy that applies to your plan.

Files also tend to carry more than you intend. A PDF can hold author names and revision history; a photo can carry location data. A little care goes a long way:

• Strip metadata from sensitive documents before uploading.
• Avoid uploading full customer databases, passports, private keys, or confidential contracts unless you fully understand the policy that applies.
• Share a redacted version whenever Claude only needs part of a document.

Is Claude Private Enough for Sensitive Work?

For a lot of everyday use, Claude's controls are enough — especially if you turn the training setting off, which keeps your conversations out of the five-year training pipeline (safety-flagged and self-reported material aside). Anthropic's business and API terms are stricter still, which is why serious or regulated work usually belongs there rather than on a personal plan.

For genuinely sensitive material — legal documents, client data, health information, unreleased work, source code — the level of care should match the stakes. That is less about Claude being "bad" and more about the simple fact that anything you paste is processed on someone else's servers. The right workflow for a public blog draft and for a confidential settlement are not the same.

How Secret Chat AI Uses Claude Differently

The most direct way to stop worrying about what Anthropic stores is to stop being identifiable to it in the first place. That is what Secret Chat AI is built for: it is an anonymizer and depersonalizer. Its whole design is no profile, no association, no training — it makes you unlinkable to the model providers, so you get the power of Claude as a stranger.

In practice, using Claude through Secret Chat AI means:

• No profile of you. Registration takes an email — used only for account access and payment, never to store or associate your prompts with you — with no name, phone number, Anthropic account, or behavioral portrait built up across your chats.
• No chat associated with you. It reaches Claude through Anthropic's business API on your behalf, so Anthropic sees the gateway, not you — no identity, no account, and no IP address attached to any prompt.
• Never used for training. Under those business-API terms your prompts are not used to train Claude, so the 2025 consumer changes (training on by default, up to five-year retention) simply do not apply.
• History only in your browser. Your conversations live in local storage on your device, not a cloud archive under your identity. You can switch between Claude and other private AI models in one place, each with a Session Privacy Report (PDF) noting the retention that applied.

This reframes retention entirely: even where Anthropic's API keeps data for a window, what is kept is not linked to you. The one honest boundary is that Secret Chat AI is an anonymizer, not a content filter. Anthropic's servers still have to read your prompt to answer it, and Secret Chat does not encrypt or strip the words you send — it removes who you are from the request, not what the request says. So if a prompt contains names or identifiers you would not want any model to see, take them out before sending.

Practical Tips: Using Claude More Privately

• Open your Privacy Settings and confirm whether the training setting is on or off — turning it off keeps new chats out of the five-year training pipeline. Then delete what you no longer need, since nothing expires on its own.
• Delete conversations you no longer need; deleted chats are excluded from future training.
• For work with real stakes, use business or API access rather than a personal plan.
• Keep passwords, private keys, full IDs, and raw customer data out of prompts entirely.
• Redact documents before uploading, and share only the part Claude needs.
• Use a private AI interface when you want local browser-based chat storage instead of a cloud history.

Table: Claude Data Retention at a Glance

Frequently Asked Questions

1. Does Claude save my conversations?

   Yes. On consumer plans your chats are saved to your account until you delete them, with incognito chats the exception. Deleting one clears it from history at once and from back-end storage within about 30 days. Separately, if you allow model improvement, a de-identified copy can sit in Anthropic's training pipelines for up to five yearsor on.

2. Does Anthropic train Claude on my chats?

   On the consumer plans (Free, Pro, Max), yes by default since the August 2025 update — unless you turn the setting off. On the business API and Claude for Work, Anthropic does not train on your data without explicit permission.

3. How long does Anthropic keep my Claude data?

   On consumer plans your history lasts until you delete it; deletion then completes on the back end within about 30 days. Allowing model improvement additionally puts a de-identified copy in a training pipeline for up to five years. Safety-flagged inputs and outputs can be kept for two years, and the classification scores for seven. On the API and commercial products, conversation content is not retained by default, though some models require a 30-day window. Safety-flagged content may be kept up to two years.

4. If I delete a Claude conversation, is it gone?

   Deleting removes it from your history, and Anthropic says deleted conversations are excluded from future training. Exact backend purge timing can vary, so check Anthropic's current documentation rather than assuming instant erasure everywhere.

5. Is Claude more private than ChatGPT?

   It depends on the setting and product, not on a blanket verdict. Since August 2025, consumer Claude trains on chats by default unless you opt out — similar in spirit to other consumer chatbots. The bigger privacy gains come from using business/API terms or a privacy-focused interface, and from keeping sensitive data out of prompts.

6. Does Secret Chat AI store my Claude chats?

   Secret Chat AI keeps your conversation history in your own browser's local storage rather than an app-side cloud archive, and it accesses Claude through Anthropic's business API, where your prompts are not used for training. Anthropic still processes each prompt to answer it, so it is not a fully private, on-device tool.

Conclusion

So, does Claude store your conversations? Yes — but the useful answer is that it depends on a choice and a product. Since August 2025, consumer Claude uses new chats to train models unless you opt out, and allowing it puts a de-identified copy in a training pipeline for up to five years. Opting out does not clear your history for you: chats stay until you delete them, and deletion completes on the back end within about 30 days. The business API and Claude for Work are stricter: no training by default, minimal retention, and options like Zero Data Retention. And in all cases, safety-flagged content can be kept longer.

Once you know which lever applies to you, the picture is far less mysterious. If you want Claude's strengths with a cleaner privacy workflow — local browser-based history, email-only sign-up, business-API access that is not used for training, and a choice of models — Secret Chat AI is built for exactly that, without pretending to remove Anthropic's processing of your prompt.

Official Anthropic Sources

• Anthropic — Updates to Consumer Terms and Privacy Policy
• Anthropic — Privacy Policy
• Anthropic — How long do you store my data? (consumer)
• Anthropic — How long do you store my organization's data? (commercial)
• Anthropic — API and data retention (ZDR & HIPAA)