Privacy and data basics
What to think about before you paste something into any AI tool.
Using Claude means sending text, and sometimes files, to a service run by another company. That is normal and often fine. It also deserves a moment of thought, because you cannot take a message back once it is sent.
We are not lawyers and this club is independent of Anthropic. This lesson gives you questions to ask. The answers about retention, training, and sharing live in Anthropic's own privacy policy and usage terms, and they differ by product and plan. Read those pages directly at anthropic.com and docs.claude.com instead of relying on a summary, including ours.
Questions to ask for any tool
- Is my content used to train models? Policies differ between consumer apps, team plans, and the API. Find the current statement for your plan and check whether there is a setting you control.
- How long is it kept? Look for retention details and whether you can delete conversations.
- Who can see it? On a work or team account, an administrator may have access. Your employer may also have rules.
- Where does it go next? If you connect other tools or services, those have their own policies.
Know your own rules first
Before any policy question, there are rules that already apply to you:
- Your employer's acceptable-use or AI policy.
- Client contracts and confidentiality agreements.
- Laws and professional rules for regulated data, such as health records, financial records, student records, or legal privilege.
If one of those forbids sharing the material with a third-party service, the answer is no regardless of the tool's settings.
Practical habits
- Minimize. Share the part you need, not the whole file.
- Redact. Replace names, emails, account numbers, and addresses with placeholders such as [CLIENT] or [ACCOUNT]. You can paste the reply back into your own document and fill the placeholders in locally.
- Never paste secrets. Passwords, API keys, private keys, and tokens do not belong in a chat. If you pasted one by accident, treat it as exposed and rotate it.
- Separate accounts. Keep personal and work use apart so a work rule never accidentally applies to a personal chat, or the other way around.
- Review settings periodically. Privacy controls can change as products change.
Accuracy is a privacy issue too
Models can state false things about real people. Do not use an answer to make claims about a named person without checking a real source. Do not ask a model to guess private facts about someone.
When using Claude Code or the API
Tools that read your project folder can see everything they are allowed to read. Keep secrets out of the folder, use ignore rules and permissions, and read the data-handling pages for the product you use. Track B covers permissions.
Try it
Take a real document you were considering sharing and make a redacted copy. Replace every name, number, and address with a placeholder. Then ask Claude to do the task on the redacted copy. Afterward, write down three things: what you removed, whether the answer was still useful, and where you found your plan's data-handling policy. If you cannot find that last one in ten minutes, that is itself a finding.