If you use Claude for anything sensitive, it is reasonable to ask what Anthropic does with your conversations. The honest answer is that it depends heavily on which Claude you are using — the consumer app, the API, or a business product — and on the settings in effect for your account. Here is how to reason about it.
Consumer Claude.ai versus the API
Anthropic draws a clear line between its consumer product and its commercial offerings. The consumer Claude.ai app, the kind individuals sign up for, is governed by a consumer privacy policy and exposes data-use settings in your account. The commercial API and business products (Claude for Work, Team, and Enterprise) are governed by separate commercial terms that historically state inputs and outputs are not used to train Anthropic’s models by default. In short, business and API traffic carry stronger contractual protections than the free-to-use consumer app, where behaviour depends on your chosen settings.
Is your data used for training?
For API and business products, the default position has been that your prompts and the model’s responses are not used to train models. For the consumer Claude.ai product, whether conversations may help improve models depends on the data settings applied to your account, and Anthropic has changed these options over time. The only reliable way to know your status is to open your account’s privacy settings and check what is currently enabled, rather than relying on what was true a year ago.
How long conversations are kept
Anthropic stores conversation data to run the service, support features, and meet safety obligations. The exact retention window varies by product and by whether content gets flagged for trust-and-safety review, in which case it may be kept longer. When you delete a conversation, it is removed from your account, though copies can persist briefly in backups or where Anthropic has a legal duty to retain them. The current numbers live in Anthropic’s privacy policy and should be checked there.
Controlling and deleting your data
You have meaningful controls. In Claude.ai, the account privacy or data settings let you review and adjust how your conversations may be used, and the chat controls let you delete individual conversations. For organisations on business or enterprise plans, an administrator configures data handling centrally according to the signed agreement. Because the available toggles evolve, the safest habit is to visit Anthropic’s Privacy Center and your account settings periodically.
Practical guidance
Treat any cloud AI assistant, Claude included, as you would email: convenient, but not the place for secrets you cannot afford to have stored. For routine work this is fine. For confidential client data, regulated information, or trade secrets, prefer the API or a business tier with explicit no-training terms, check your data settings, and avoid pasting material you would not want retained. Anthropic’s defaults are relatively privacy-friendly compared with some peers, but defaults are no substitute for reading the current policy and setting your own controls.