Claude Code Privacy: What Your Account Settings Control
Yes—Claude Code privacy depends on which account or organization you use and the settings and policies that apply to it. Your plan can affect whether coding sessions may be used to improve models, while organizational controls, product features, and the data you choose to share also matter.
What Claude Code can access
Claude Code is a coding assistant that works with a project through a command-line interface. Depending on how you start it and the permissions you grant, it can inspect files, suggest or make code changes, and run commands. That means privacy is not determined by one “Claude Code” switch: it involves both the information sent to the service and what the local tool can see or do.
For a typical interactive session, the service needs relevant context to respond. This can include prompts, code or file contents you provide or that the tool reads, and the results of commands or tool actions. The exact information involved varies with your workflow and the features you use. Don’t assume that only the question you typed is transmitted if you have asked the tool to inspect a repository or execute a task.
Local access and service-side data handling are different issues. Limiting a session to a particular project or approving actions carefully can reduce unintended local exposure, but it does not by itself change how information submitted to the service is handled under your account’s terms and settings.
How account type and settings affect data use
Claude Code is available through different account arrangements, including individual accounts and organization-managed access. The applicable privacy terms and controls can differ. Consumer account settings may include a choice about whether conversations are used to improve models; business or organization accounts are generally governed by their commercial terms and administrator controls. Check the current terms for the specific account you are using rather than applying one account’s policy to another.
For individual use, look in the account’s privacy or data controls for the current model-improvement setting. Its label, location, and effect can change, so verify the wording shown in your account. Changing a setting should not be treated as a way to erase information already retained or as a guarantee that every form of operational processing stops.
If you use an organization account, your administrator may manage access, retention, or other data controls. A company policy can impose rules beyond the options visible to an individual user. If the account is connected through an organization, ask the administrator which terms and controls apply, especially before using it with confidential code.
- Account type: individual, team, or other organization access may have different governing terms.
- Data-use controls: review whether submitted content may be used for model improvement.
- Retention and administration: check the applicable terms and organization policy; do not infer these from a single toggle.
- Features used: integrations, tools, and command execution can change what information enters a session.
What “not used for training” does—and does not—mean
A common misconception is that disabling model improvement makes a session private in every sense. It addresses a particular use of data; it does not necessarily mean that no data is processed, that nothing is retained for service operation or safety, or that your employer cannot apply its own logging and governance. The applicable product terms explain these separate purposes and any retention rules.
Likewise, using an account covered by business terms should not be taken to mean that you can submit any secret without checking your organization’s policy. The service still has to process information to provide responses, and an organization may have its own requirements for source code, customer data, credentials, or regulated information. Privacy is a combination of service terms, settings, access choices, and what you send.
Ways to reduce accidental exposure
Use the least sensitive context that will solve the task. Before starting a session, consider what is in the repository and what commands or tools the assistant may use. These practical steps reduce avoidable exposure, but they do not replace your organization’s security rules or the service’s terms.
- Review the account and organization policy before using Claude Code on work or proprietary projects.
- Do not paste passwords, access tokens, private keys, or other credentials into prompts. Keep secrets out of files and logs that could be included as context.
- Limit the working directory to the project needed for the task, and avoid granting broad access to unrelated files.
- Read proposed commands and changes before approving them, particularly commands that access files, contact external services, or modify a large part of a project.
- Use appropriate repository exclusions or configuration where supported, but verify their behavior rather than assuming a sensitive file is automatically ignored.
- For confidential work, use only an account and workflow approved by the organization responsible for that code.
Choosing the right account and workflow
For personal experiments with non-sensitive code, an individual account may be sufficient after you review its current data controls. For work repositories, the key question is not simply whether Claude Code works technically; it is whether your organization has approved the account, terms, and workflow. A personal account may not carry the protections or administration your employer requires.
Before choosing, answer these questions:
- Is the project public, personal, proprietary, or subject to a confidentiality requirement?
- Which account is active, and what data-use terms apply to it?
- Does your organization permit code to be submitted to this service and, if so, under which account?
- Can you narrow the files and actions available to the session?
- Are you relying on a setting to delete or exclude data when the terms do not promise that result?
If you cannot establish the applicable policy, pause before using the tool with sensitive material and ask the account owner or administrator. This is more reliable than assuming that a privacy option on one account applies to another.
Frequently Asked Questions
Does Claude Code send my whole repository?
Not necessarily. What is sent depends on what you ask it to inspect, the files and context involved, and the features used. Avoid assuming that repository access means only a small prompt is processed; keep the working scope narrow and review the applicable product behavior.
Can I use Claude Code with private company code?
Only if your organization’s rules and the account’s terms permit it. Confirm the approved account and workflow with your administrator before submitting proprietary code.
Does turning off model improvement stop all data retention?
No. That setting concerns model improvement, not every possible form of processing or retention. Consult the current terms and privacy controls for the account to understand what applies.
Does local command execution mean my code stays on my computer?
No. Local execution describes where commands run; it does not establish that prompts or code context used to generate a response never leave the computer. Consider the service-side data terms separately.
Are personal and organization accounts handled the same way?
Not necessarily. Their terms, controls, and administrative arrangements can differ. Check the policy for the account actually signed in, and ask an organization administrator when applicable.
Related
Method: the standard formula for this calculation, rounded as each result states. See how the calculators work or report an error.