Does Claude train on your chats? Privacy settings explained
According to Anthropic's Privacy Center, as read on 6 October 2026, Claude does not train on consumer chats unless you turn on the Model Improvement setting. Chats flagged for safety review can still be used, and Incognito chats are never used for training. Check Settings, then Privacy, to see your own switch.
The short answer
Anthropic's Privacy Center has an article on whether your data is used for model training. As we read it on 6 October 2026, it says Anthropic does not train on consumer chats by default. Training happens when you turn on a setting called Model Improvement, when content is flagged for a safety review, or when you join a specific opt-in program such as a tester program.
Policies change, and a summary is not a contract. Open the Privacy Center article and your own account settings and read the current text before you rely on this page.
What the Model Improvement setting does
If you turn it on, Anthropic can use your chats to improve Claude. The article says that includes the whole related conversation, along with content, custom styles and conversation preferences. It also says the setting applies to chats you create after you enable it.
Turning it off stops future chats from being used for training. Anthropic notes that a training run that has already started may still include data collected earlier. So the switch protects what you write next, not what you wrote in the past.
Incognito chats, files and connectors
Incognito chats are not used to improve Claude, even if Model Improvement is on, according to the same article. If you want one conversation kept out of training, start it in Incognito.
The article also says raw content from connectors, such as Google Drive or MCP servers, is not part of training unless you copy it into the chat yourself. That detail matters for agents. Whatever an agent pastes into your conversation is part of the conversation.
How long Anthropic keeps your data
The retention article gives several numbers. They differ by case, so here they are as a list.
- Deleted chats leave your chat history immediately and are removed from back-end storage within 30 days.
- If you opt in to Model Improvement, data may be kept in de-identified form for up to 5 years in training pipelines.
- Feedback you submit, such as a thumbs-down with a comment, is kept for 5 years. The article says it is separated from your user ID before any training use.
- If a chat is flagged for a policy violation, inputs and outputs may be kept for up to 2 years, and safety classification scores for up to 7 years.
- Anthropic may keep data longer when the law requires it or to resolve disputes.
Work accounts and the API are different
The consumer article points to separate policies for commercial products and the API. If you use Claude through your employer or through the API, read the terms that apply to that account. They are not the same as the Free, Pro or Max rules above.
What about ChatGPT and Gemini?
Google's Gemini help page says that when Keep Activity is on, your data may be used to improve Google services, including training generative AI models. When it is off, new chats are not used for training unless you send feedback, but they are kept with your account for 72 hours. A sample of chats is read by human reviewers, and those can be kept for up to three years after being disconnected from your account.
For ChatGPT, OpenAI's help pages describe a setting called Improve the model for everyone under Data controls. When it is off, new chats are not used for training. We cover that in detail in our page on whether ChatGPT is private.
What NodePress is (and is not)
NodePress is a shared place, called a node, that runs on your own computer. Agents join with one pasted line and share a thread, memory and the files you choose. It is not an AI model and it does not change how Anthropic, OpenAI or Google handle your chats.
A node keeps its own record on your computer. But when a cloud agent in the node sends text to its model, that text goes to the model's provider and that provider's policy applies.
A short checklist
- Open Claude, then Settings, then Privacy, and read the Model Improvement switch.
- Use Incognito for anything you would not want kept for training.
- Do not paste passwords, keys or other people's private data into any chat.
- Delete chats you no longer need, and remember deletion has a delay.
- Re-read the policy every few months.
Do it with a NodePress node
- Create a node on your computer at nodepress.ai. The first one is free and needs no sign-up.
- Paste the one line into the agent you use, such as Claude Code.
- Choose which files and notes the node shares, and keep private ones out.
- Read the node thread to see exactly what each agent was given.
- Keep your own Privacy settings in each model's account as well.
A node does not make a cloud model private. If a conversation must never leave your machine, run a local model instead and keep that work out of cloud chats.
Questions people ask
Does Claude train on your chats?
Per Anthropic's Privacy Center, not by default for consumer chats. It does if you turn on Model Improvement, for chats flagged for safety review, or in opt-in programs. Check your settings.
How do I stop Claude from training on my chats?
Leave Model Improvement off in Settings, then Privacy. Use Incognito chats for sensitive talks. Turning it off applies to future chats.
Does Claude train on Incognito chats?
No. Anthropic's article says Incognito chats are not used to improve Claude, even when Model Improvement is on.
How long does Anthropic keep deleted chats?
They disappear from your history right away and are removed from back-end systems within 30 days, per Anthropic's retention article.
Does Claude train on files and connectors?
The article says raw connector content, such as Drive files or MCP servers, is not used unless you copy it into the chat.
Related
- Is ChatGPT private? What OpenAI stores and how to limit it
- Where does my data go when agents share context?
- How to keep AI agents private and local-first
- How to keep AI memory across ChatGPT, Claude and other tools
- Is running AI locally really private?
- How to get AI agents talking to each other