Privacy Mode and data handling

To answer a request, Cursor sends relevant code, your prompt and context to its servers, which build the final prompt and call the model provider. That is true even when you use your own API key: requests still pass through Cursor's backend.

Privacy Mode controls what happens to that data afterwards. With it on, Cursor states that your data is not used for training, and it relies on zero data retention agreements with model providers, so providers don't store or train on it. Recent releases name the options "Privacy Mode (no training)" and "Legacy (no storage)"; check the current wording in settings and on the data-use page. Some models are not covered by zero-retention agreements; they are labeled as such, and on teams an admin has to opt in before members can use them.

With Privacy Mode off, Cursor may store and use prompts, code snippets, codebase data and editor actions to improve its features and train models.

Other details worth knowing:

• File contents may be temporarily cached on Cursor's servers to reduce latency, encrypted with keys that exist there only during a request.
• Abuse-detection systems are an exception: flagged content may be retained for investigation.
• Team admins can enforce settings for everyone.

Practical guidance: turn Privacy Mode on for work code (or confirm your organization enforces it), follow your company's policy about which tools may see which code, keep secrets out of files the AI can read, and read the current data-use page yourself rather than relying on memory; these policies change.