Does Anthropic Train on Your Data? The Answer for Every Claude Plan

Does Anthropic train on your data? Yes, if you use Claude Free, Pro or Max and leave model training on: under Anthropic's consumer terms, in force since October 8, 2025, those chats and coding sessions can train Claude and be kept for up to 5 years. No, if you use Claude for Work (Team or Enterprise), Education, Government or the API, including through Amazon Bedrock and Google Vertex AI: Anthropic's commercial terms say it "may not train models on Customer Content from Services."

To opt out on a personal plan, open Settings, then Privacy, and switch off Help Improve our AI models. Incognito chats are never used for training. Two things can still be used after you opt out: chats you rate with a thumbs up or down, and conversations Anthropic's safety systems flag.

After nearly 20 years in AI development and digital marketing, I've read more AI privacy policies than I'd like to admit, and Anthropic's changed more in 2025 than most do in five years. This post goes plan by plan: what Anthropic trains on, how long it keeps your chats, what survives an opt out, and the exact setting to change.

My Main Points:

• Claude Free, Pro and Max chats and coding sessions train Claude by default. Anthropic announced the change on August 28, 2025, and the new consumer terms took effect on October 8, 2025.
• With training on, Anthropic can keep your data for up to 5 years. With it off, Anthropic's August 28, 2025 update post kept the existing 30 day retention period.
• Claude Team, Enterprise, Education, Government and the API are not trained on by default. Anthropic's privacy center wording: "By default, we will not use your inputs or outputs from our commercial products" to train models.
• Three things sit outside the toggle: feedback you send with a thumbs up or down (kept up to 5 years and usable for training), chats flagged by safety systems (inputs and outputs kept up to 2 years, classifier scores up to 7 years), and incognito chats, which are never used for training.
• The consumer switch lives under Settings, then Privacy, and is called Help Improve our AI models. Team and Enterprise owners control the feedback path with the Rate chats setting.

In this guide, I'll show you the rules for each plan, including Claude Code on a personal account, and how Anthropic's policy compares with OpenAI's.

Does Anthropic Train on Your Data? The Answer by Claude Plan

Anthropic runs two rulebooks. Personal plans fall under the consumer terms, where training is on unless you opt out. Everything a company or developer pays for falls under the commercial terms, where training is off unless you hand data over through feedback or a separate agreement.

Claude plan	Trained on by default?	How long Anthropic keeps chats	Where you control it
Claude Free, Pro and Max (including Claude Code on those plans)	Yes, unless you opt out	Up to 5 years with training on; 30 day retention with it off; deleted chats removed within 30 days	Settings, then Privacy: Help Improve our AI models
Claude Team and Enterprise (Claude for Work), Education, Government	No	Kept in the product until you delete them; deleted chats removed within 30 days	Organization settings, then Data and Privacy: Rate chats (feedback)
Claude API, including Amazon Bedrock and Google Vertex AI	No	Privacy center: deleted within 30 days. API docs: not retained by default, except Covered Models (30 days)	Zero data retention agreement, on request

Claude Free, Pro and Max: Trained Unless You Opt Out

Claude Free, Pro and Max users give Anthropic permission to train on new or resumed chats and coding sessions unless they switch training off. Anthropic announced the change in its August 28, 2025 update post, gave existing users until October 8, 2025 to pick a setting, and put the new consumer terms into force on that date. The terms say Anthropic may use your content "including training our models, unless you opt out of training."

The change covers Claude Code too, when you run it on a Free, Pro or Max login. A coding session on a personal plan is consumer data; the same session through an API key from a commercial organization is not.

Retention follows the setting. With training on, Anthropic's privacy center says it may keep your data, stripped of identifiers, for up to 5 years in its training pipelines. With it off, the August 2025 update post kept the existing 30 day retention period, and deleted chats leave Anthropic's back end within 30 days either way.

Switching training off applies to the future. Anthropic says it will not use your previous or new chats for future training once the setting is off, and deleted chats are not used either, but data already inside a training run that has started, or inside a model that has shipped, stays there.

Claude Team, Enterprise, Education and Government: Not Trained by Default

Claude Team and Enterprise (sold together as Claude for Work), Claude for Education and Claude Gov sit under Anthropic's commercial terms, which say Anthropic "may not train models on Customer Content from Services." Anthropic's commercial privacy page puts it in plainer words: "By default, we will not use your inputs or outputs from our commercial products" to train models, and it names Claude for Work, the API and Claude Gov as examples.

The words "by default" leave one door open, and it is feedback. If a team member rates a reply with a thumbs up or thumbs down, Anthropic stores the whole conversation for up to 5 years and may use it for training, after separating it from user and customer IDs. A Primary Owner or Owner of a Team or Enterprise plan can close that path with the Rate chats setting under Organization settings, then Data and Privacy.

The August 2025 consumer change did not touch these plans. Anthropic's update post names Claude for Work, Claude for Government, Claude for Education and API use as outside its scope.

Claude API, Amazon Bedrock and Google Vertex AI: Not Trained by Default

Claude API traffic is commercial data, so the same "may not train" clause applies, and Anthropic's API data retention docs add that retained data "is never used for model training without your express permission." Anthropic's update post also lists access through Amazon Bedrock and Google Cloud's Vertex AI as outside the consumer change; on those platforms the cloud provider, not Anthropic, is the data processor.

Anthropic's own pages disagree on API retention. The commercial privacy center page, updated July 1, 2026, says Anthropic deletes API inputs and outputs on its back end within 30 days. The API docs say conversation content "is not retained by default," with one exception: Covered Models, which require 30 day retention. If 30 days is too long, a zero data retention agreement is available through Anthropic's sales team, and even under that agreement flagged content can be kept for up to 2 years.

What Anthropic Still Uses After You Opt Out

Anthropic keeps three paths open regardless of your training setting, and it helps to know each one before you assume you are fully opted out.

• Feedback. A thumbs up or thumbs down sends the entire conversation to Anthropic, which keeps it for up to 5 years and may use it for training even if you opted out. Skip the rating buttons on anything you would not want in a training set.
• Safety flags. If Anthropic's automated trust and safety systems flag a chat as a Usage Policy violation, Anthropic keeps the inputs and outputs for up to 2 years and the classifier scores for up to 7 years. This applies on consumer and commercial plans alike, and even under a zero data retention agreement.
• Legal holds. Anthropic may keep chats as required by law, to resolve disputes, or to enforce its Usage Policy.

Incognito chats go the other way: Anthropic says they "are not used to improve Claude, even if you have enabled Model Improvement" in your privacy settings. Incognito is the quickest option for a single sensitive chat when you do not want to touch the global setting.

How to Opt Out of Claude Training

Opting out of Claude training on a Free, Pro or Max account takes four clicks, and Anthropic's settings guide confirms the path on desktop and mobile:

1. Select your name in the menu and open Settings.
2. Open the Privacy tab.
3. Find Help Improve our AI models.
4. Toggle it off.

The same choice appears in the signup flow for new users, and it appeared in a pop up window for existing users before the October 8, 2025 deadline. If you clicked through that window without reading it, training is probably on, and checking takes a minute.

The switch does not delete anything, and it does not pull your data out of a training run that already started. To remove a chat, delete it; Anthropic removes deleted conversations from its back end within 30 days. To keep one conversation out of training without changing the global setting, start it in incognito mode.

On a Team or Enterprise plan there is no training toggle because training is already off by default. The one setting worth checking is Rate chats under Organization settings, then Data and Privacy, which an owner can turn off to stop members from submitting feedback. On the API, the control you have is retention: ask Anthropic's sales team for a zero data retention agreement if your workload needs it.

Anthropic Data Policy Timeline

Anthropic's data policy shifted in three steps in 2025, and most of the confusion I see online comes from mixing them up.

• June 17, 2025: the current commercial terms take effect, including the clause that Anthropic "may not train models on Customer Content from Services."
• August 28, 2025: Anthropic publishes its update post. Free, Pro and Max users are asked to choose whether chats can train Claude, and retention extends to 5 years for those who allow it.
• October 8, 2025: deadline for existing users to make a choice, and the date the new consumer terms take effect. From this point, training is on unless you opt out.
• July 1 and August 18, 2026: Anthropic's privacy center refreshes its retention and training pages, which is where the current 30 day, 2 year, 5 year and 7 year figures come from.

Anthropic vs OpenAI on Training Data

Anthropic and OpenAI now run the same basic model for personal accounts: training is on unless you turn it off. ChatGPT trains on consumer chats and uploads when the "Improve the model for everyone" setting is on, per OpenAI's help center, and OpenAI's enterprise privacy page says Business, Enterprise, Edu and API data is not used for training by default, with API inputs kept for up to 30 days and zero data retention available. Those are the same three tiers Anthropic uses.

Anthropic ties retention to the training setting (up to 5 years on, 30 days off), while OpenAI keeps saved chats until you delete them whatever the setting says. OpenAI's Temporary Chat and Anthropic's incognito mode play the same role, since neither is used for training. I cover the ChatGPT side, including photos and the Library, in does ChatGPT save your data.

FAQ

Does Anthropic train on Claude Team or Enterprise data?

No, not by default. Anthropic's commercial terms say it "may not train models on Customer Content from Services," and its privacy center adds: "By default, we will not use your inputs or outputs from our commercial products" to train models. The one route around that is a thumbs up or thumbs down rating, which a Team or Enterprise owner can switch off with the Rate chats setting.

Does Anthropic train on API data?

No, not by default. API traffic falls under the same commercial terms, and Anthropic's API docs say retained data "is never used for model training without your express permission." Retention is less clear: the privacy center says API inputs and outputs are deleted within 30 days, while the API docs say conversation content is not retained by default except for Covered Models, which require 30 day retention.

Does Claude train on your data if you turn off Help Improve our AI models?

Mostly no. With the setting off, Anthropic says it will not use your previous or new chats and coding sessions for future training. Two exceptions remain: conversations you rate with a thumbs up or down can still be used, and chats flagged by Anthropic's safety systems can be kept for up to 2 years. Data already inside a training run that started before you opted out stays there.

Are Claude incognito chats used for training?

No. Anthropic's privacy center says incognito chats "are not used to improve Claude, even if you have enabled Model Improvement" in your privacy settings, so incognito is the quickest way to keep a single sensitive conversation out of training.

How long does Anthropic keep Claude chats?

On Claude Free, Pro and Max with training on, Anthropic can keep your data, stripped of identifiers, for up to 5 years. With training off, Anthropic's August 28, 2025 update post kept the existing 30 day retention period. Deleted chats leave Anthropic's back end within 30 days on every plan, and safety flagged chats can be kept for up to 2 years with classifier scores kept up to 7 years.

Final Thoughts

Does Anthropic train on your data? On a personal Claude plan, yes until you say otherwise, and the toggle has been on by default since October 8, 2025. On Team, Enterprise, Education, Government and the API, no by default, with feedback and safety flags as the two routes around that. Check Settings, then Privacy once, skip the thumbs buttons on anything sensitive, and use incognito for the rest.

I still rate Anthropic's setup above most of the field: one visible switch, retention that follows it, and official wording you can quote to a compliance team. If your company wants Claude or another model running with its own data controls, that is the kind of deployment we build through our AI development services, and our guide to private AI explains when it makes sense.