How safe and private is it to share sensitive information with Claude?

For claude.ai free and Pro users, Anthropic may use conversations to improve its models by default. You can opt out of this in your account privacy settings under Data Controls. When you opt out, your conversations are not used for training. Always check the current privacy policy at anthropic.com as these settings and policies can change.

If you are using Claude for business or enterprise work, Anthropic offers a Claude for Work plan and an API with a clear data retention policy. By default, API inputs and outputs are not used to train models, which is the standard expectation for business API usage. Check your API usage agreement for the exact terms.

For highly sensitive data like patient records, legal privileged information, or financial data under regulatory requirements, the safest approach is to use the Claude API through a private cloud setup such as AWS Bedrock with Claude. This way your data stays within your own cloud infrastructure and is subject to your own compliance controls rather than passing through Anthropic's consumer systems.

A practical middle ground for work use is to anonymize or redact truly sensitive identifiers before pasting into Claude. Replace real names with Person A and Person B, use placeholder company names, and remove account numbers or IDs. Claude can still help you reason about the substance of the content without needing the real identifying details.

Claude has a feature called Incognito conversations in claude.ai where your chats are not saved to your history and are not used for training. This is useful for particularly sensitive one-off queries. Look for the option in the new conversation menu. Note that even in incognito mode the data still passes through Anthropic's servers — it is just not stored afterward.