Does Claude train on your data? Yes on Free, Pro and Max if Model Improvement is on; no by default on Team, Enterprise and the API

Does Claude train on your data? It depends on your plan and one setting. On Claude Free, Pro and Max, Anthropic uses your chats and Claude Code sessions to train its models when the Model Improvement setting is on, and keeps that data for up to five years. On Team, Enterprise, Claude for Government, Claude for Education and the API, it doesn't train on your content by default. The question drew fresh attention on October 27, 2026, when Anthropic's privacy center page "Is my data used for model training?" reached 217 points on Hacker News. For what each plan costs and includes, see our guide to the Claude app and its plans.

This explainer is based on Anthropic's Privacy Center articles, its Consumer Terms (effective October 8, 2025), Commercial Terms, Privacy Policy (effective September 10, 2026), its August 28, 2025 consumer terms announcement, the Claude Code data usage docs and Claude help center pages, checked October 28, 2026.

The short version

• Free, Pro and Max: trained on if Model Improvement is on. Turn it off in Settings > Privacy. This includes Claude Code used from these accounts.
• Even with it off, Anthropic's Consumer Terms say it will still train on conversations you give feedback on (thumbs up or down) and on chats flagged for safety review.
• Team, Enterprise, Government, Education and the API: not trained on by default under the Commercial Terms. Feedback and opt-in programs are the exceptions.
• Incognito chats are never used for training, on any plan, and are kept for 30 days by default.
• Retention: up to five years if consumers allow training, 30 days if they don't, 30 days for API data, and up to two years for flagged chats.
• The change: Anthropic introduced consumer training on August 28, 2025; existing users had to choose by October 8, 2025, according to Anthropic's post.

Does Claude train on my data on the Free, Pro or Max plan?

Yes, if you've allowed it. Anthropic's privacy center says that for "consumer products such as Claude Free, Pro, Max," including "when accounts from those plans use Claude Code," it will use your chats and coding sessions "(including to improve our models)" if you choose to allow it, if a conversation is "flagged for safety review," or if you opt in some other way, such as joining its Trusted Tester Program.

The Consumer Terms frame it the other way round. They say Anthropic may use your materials for "training our models, unless you opt out of training through your account settings." Either way, the switch is the same setting.

Anthropic's announcement on August 28, 2025, said new users pick a preference when they sign up and existing users saw a pop-up. When that pop-up appeared, TechCrunch reported that it showed a large "Accept" button with "a much tinier toggle switch" for training that was "automatically set to 'On.'" Anthropic's post doesn't say how the toggle was set. Free-plan details are in our guide to what Claude's free plan includes.

How do I turn off Claude training on my data?

Use the Model Improvement setting. Anthropic's help article gives these steps:

1. On the web or desktop, open the menu under your name and select Settings.
2. Select Privacy, or go to claude.ai/settings/data-privacy-controls.
3. Under "Help Improve our AI models," switch the toggle off.

The mobile app has the same path: your name, Settings, Privacy. Anthropic says you can change the choice at any time.

Turning it off stops future use, not past use. Anthropic's privacy center says it "will not use your previous or new chats or coding sessions for future model training," but that your data "will still be included in model training runs that are already in progress, or in models that have been trained." Deleting a chat also keeps it out of future training.

For a single conversation you don't want used, start an incognito chat with the ghost icon. Anthropic's help center says "Incognito chats are not used for training," and its privacy center says this holds "even if you have enabled Model Improvement." Incognito isn't available inside projects. How incognito differs from Claude's memory feature is covered in our explainer on context windows, memory and RAG.

What still gets used if I opt out?

Two things, according to the Consumer Terms. Even if you opt out, Anthropic says it will train on materials when "you provide Feedback to us regarding any Materials" and when "your Materials are flagged for safety review to improve our ability to detect harmful content, enforce our policies, or advance our safety research."

Feedback means the thumbs up and thumbs down buttons. Anthropic says it stores "the entire related conversation" with your feedback for up to five years, de-links it from your user ID before use, and doesn't include raw content from connectors such as Google Drive, although text you paste into a chat can be included.

Does Claude train on Team or Enterprise data?

Not by default. Anthropic's Commercial Terms, which cover Claude for Work (Team and Enterprise), Claude for Government, Claude for Education and the API, say: "Anthropic may not train models on Customer Content from Services." The privacy center's commercial article, dated August 18, 2026, says: "By default, we will not use your inputs or outputs from our commercial products" for training.

The exception is feedback. If someone in the organization rates a chat or otherwise allows use of data, Anthropic says it "may use your chats and coding sessions to train our models," storing the related conversation for up to five years. Primary Owners and Owners can stop members from giving feedback with the Rate chats setting under Organization settings > Data and Privacy.

Does Claude train on API data?

No, by default. The API is under the Commercial Terms. The Claude Code docs describe one opt-in: an organization admin can enroll in the Development Partner Program, which lets Anthropic train on the materials provided. The docs say it's available "only for Anthropic first-party API," not for Amazon Bedrock or Google Cloud users.

Does Claude Code train on my code?

It follows your account. Anthropic's Claude Code docs say it will "train new models using data from Free, Pro, and Max accounts when this setting is on (including when you use Claude Code from these accounts)." For commercial users, including Team, Enterprise, the API, third-party platforms and Claude Gov, "Anthropic does not train generative models using code or prompts sent to Claude Code under commercial terms," unless the customer opts in.

Two Claude Code features send data even so. Transcripts sent with /feedback, /bug or /share are "retained for 5 years." The optional "Can Anthropic look at your session transcript" prompt after a session rating uploads the transcript only if you pick Yes; those transcripts are kept for up to six months and, the docs say, "cannot be used to train our AI models." Setting CLAUDE_CODE_DISABLE_FEEDBACK_SURVEY=1 turns the surveys off.

Claude training and retention by plan

Last verified: October 28, 2026

How long does Anthropic keep my Claude chats?

For Free, Pro and Max, the answer depends on the training choice. Anthropic's August 28, 2025 post said: "We are also extending data retention to five years, if you allow us to use your data for model training." If you don't, it said, "you will continue with our existing 30-day data retention period." The Claude Code docs give the same pair: a "5-year retention period" if you allow training and a "30-day retention period" if you don't. The privacy center adds that data kept for training is stored "in a de-identified format."

Deleting a conversation removes it from your history immediately and from Anthropic's back end "within 30 days," the Privacy Policy says. Chats flagged for breaking the Usage Policy are the exception on every plan: Anthropic keeps "inputs and outputs for up to 2 years and trust and safety classification scores for up to 7 years."

For commercial customers, the API default is that Anthropic deletes "inputs and outputs on our backend within 30 days of receipt or generation." In the Claude apps, Team and Enterprise chats stay in the product so you can return to them. Enterprise Owners can set a custom period; Anthropic's help center says data is otherwise "retained indefinitely," and "the minimum retention period is 30 days."

When did Claude start training on user data?

On August 28, 2025, Anthropic announced updated Consumer Terms and a Privacy Policy that, for the first time, let it train on Free, Pro and Max conversations. Before that, TechCrunch reported, consumer prompts and outputs were deleted from Anthropic's back end within 30 days. The changes, Anthropic said, "apply only to new or resumed chats and coding sessions," so older chats you never reopened aren't used.

The deadline is reported two ways. Anthropic's post now says existing users had "until October 8, 2025" to decide, and the current Consumer Terms carry that effective date. TechCrunch's August 28, 2025 report gave September 28. After the deadline, Anthropic said, users had to choose to keep using Claude. Anthropic also said: "We do not sell users' data to third parties."

How Claude's settings compare with ChatGPT's and Gemini's is in our Claude vs ChatGPT vs Gemini comparison.

What's confirmed, reported and uncertain

• Confirmed (Anthropic's own pages): the consumer training rule and its exceptions, the Model Improvement steps, the commercial no-training clause, the Rate chats control, the Development Partner Program, the Claude Code rules, incognito handling, and every retention period in the table.
• Reported: that the 2025 pop-up's training toggle was "On" by default (TechCrunch); the September 28, 2025 deadline (TechCrunch), which differs from the October 8, 2025 date on Anthropic's post.
• Uncertain: how the training choice is preset for someone signing up today; Anthropic's pages say users select a preference but don't say which option is preselected. The privacy center's standard retention section for consumers is currently blank, so the 30-day figure comes from the 2025 post and the Claude Code docs.