To limit whether an AI coding assistant uses your work to improve models, first identify the exact product, account tier, and selected model, then change the training or model-improvement setting that applies to that account. That setting does not necessarily stop the assistant from sending prompts and code context to a provider, retaining data, collecting telemetry, or processing feedback. Review those separately before using sensitive code.
Before changing a setting, identify which assistant and account you use
Privacy rules can differ between a personal subscription, a managed work account, an IDE extension, a command-line tool, an API, and the model selected inside an assistant. Write down the product surface, plan or workspace, model, and whether you supplied your own API key. A promise for one plan or model is not a promise for every provider available in the same editor.
As an Amazon Associate I earn from qualifying purchases.
Then open the vendor’s official privacy or data-controls documentation and check the account or workspace you actually use. Look for separate controls or terms for model training, prompts and code sent for inference, retention, telemetry, feedback, safety review, and administrator policy. A training opt-out addresses training use; it is not a “no data leaves my computer” switch.
Free tools Windows power users keep installed
One-click scans. No signup required.
What the documented controls cover
| Product and scope | Training or model-improvement rule | Other data-flow details |
|---|---|---|
| Gemini Code Assist Standard and Enterprise | Google says customer data is not used to train models without permission. | Prompts, responses, conversation history, code snippets, and cursor location are Customer Data. Prompts and responses are not stored in Google Cloud by default; customers can configure Cloud Logging to store inputs and responses. Service Data, including telemetry, is handled separately. Google Cloud’s security and privacy documentation. |
| Cursor | Privacy Mode is intended to prevent code from being used for training by Cursor or model providers. | AI features send prompts and code context to model providers. Personal API keys follow the provider’s privacy terms; some models require provider retention and are outside Cursor’s zero-data-retention agreements. Cursor’s privacy documentation. |
| GitHub Copilot | Individual subscribers can manage training use of Copilot interaction data in account settings. GitHub says Copilot Business and Enterprise customer data is not used by GitHub to train AI models. | Interaction data can include prompts, suggestions, and code snippets. Handling also depends on the selected model and hosting arrangement. GitHub Copilot and GitHub’s model-hosting reference. |
| Claude Free, Pro, and Max, including Claude Code used with those accounts | Consumer chats and coding sessions may be used for model improvement if you opt in, if they are flagged for safety review (for safety purposes), or through another explicit opt-in. Incognito chats are not used to improve Claude, even when Model Improvement is enabled. | Feedback and retention have separate rules; commercial users are covered by separate terms. Anthropic’s model-training explanation and retention information. |
Set the controls for your product
Cursor: enable Privacy Mode
- Open Settings → General → Privacy Mode. Cursor documents Cmd+Ctrl+Shift+J on Mac and Ctrl+Shift+J on Windows and Linux as shortcuts to open the relevant settings.
- Turn on Privacy Mode. Cursor says this prevents code from being used for training, but prompts and code context are still sent to model providers when you use AI features.
- If you use a personal API key or a model that is disabled by default, check the provider’s terms and your organization’s approval before sending code. Cursor says some models require provider retention and sit outside its zero-data-retention agreements.
- If this is a managed account, ask whether an administrator enforces Privacy Mode and restricts available models. Cursor documents organization-wide enforcement, model restrictions, agent permissions, and audit logs.
GitHub Copilot: check the setting for your account type
- For an individual subscription, open your GitHub account settings and locate the Copilot setting for use of interaction data in model training. Use the settings link in GitHub’s Copilot documentation for the current instructions; the cited documentation does not establish a stable click-by-click path.
- Review the model you select as well as your plan. GitHub’s model-hosting reference describes provider and hosting differences, so do not assume one provider’s handling applies to every model in Copilot.
- For Copilot Business or Enterprise, confirm that you are signed into the managed workspace and that its applicable policy covers the model and features you intend to use. GitHub states that it does not use Business or Enterprise customer data to train AI models.
Claude and Claude Code: distinguish model improvement, Incognito, and feedback
- If you use Claude Free, Pro, or Max, including Claude Code with one of those accounts, review Anthropic’s model-improvement explanation and the Model Improvement control in Privacy Settings. Choose whether to opt in; do not assume a Claude consumer setting governs a commercial account or API use.
- For a conversation you do not want used to improve Claude, use an Incognito chat where available. Anthropic says Incognito chats are excluded from model improvement even if the account’s Model Improvement setting is enabled.
- Before sending thumbs-up or thumbs-down feedback, consider that it may include the related conversation. Anthropic says feedback may be retained for up to five years.
Gemini Code Assist Standard and Enterprise: distinguish customer data from service data
Google’s documentation describes the service’s training commitment separately from storage and telemetry. For these editions, prompts and responses are not stored in Google Cloud by default; a customer can configure Cloud Logging to store inputs and responses. Listed Service Data telemetry includes events such as a request being made or a response received without request contents, user reactions, accepted-suggestion character counts, and UI interactions. Review your organization’s logging configuration and IAM access management rather than treating the training commitment as a control over all data collection. Google says processing generally occurs near the request origin, but regionality is not guaranteed. These statements apply to Gemini Code Assist Standard and Enterprise, not automatically to every Gemini product or account tier.
#1 Best Overall
Review retention, telemetry, and provider processing separately
Before using the assistant with private code, answer these questions for the product, plan, and model you identified:
- Inference: What prompts, snippets, open or adjacent files, conversation history, or other context are sent to answer a request? Cursor, for example, says AI features send prompts and code context to model providers.
- Retention and logging: Are prompts and responses stored by default, and can a workspace administrator enable logging? For Gemini Code Assist Standard and Enterprise, Google documents default non-storage of prompts and responses in Google Cloud and optional Cloud Logging.
- Telemetry: Does the service collect usage or interaction signals separately from prompt contents? Check the service’s definition rather than assuming “not used for training” means “not collected.”
- Feedback and review: Does submitting a rating include the conversation? Are safety-flagged sessions handled under a separate rule? Anthropic’s consumer documentation distinguishes model-improvement use, safety review, feedback, and retention.
- Provider and key: Does the selected model use a different provider, or are you supplying your own API key? Cursor says personal API keys are governed by the provider’s privacy policy, and some models are outside its zero-data-retention agreements.
Do not paste credentials, secrets, regulated information, or proprietary code until your employer’s policy and the applicable provider terms allow that use. A training control cannot substitute for permission to send data for inference.
For a work account, have an administrator verify enforcement
Ask the administrator to confirm the workspace’s applicable agreement and the exact policies in force, not just the setting visible in your personal profile. Where supported, verify model allowlists or restrictions, organization-wide privacy enforcement, agent permissions, IAM access, and audit or logging configuration. Cursor documents organization-wide Privacy Mode enforcement and model and agent controls; Gemini Code Assist supports IAM access management. For Copilot, confirm the Business or Enterprise workspace and the selected model’s hosting arrangement.
After the policy is set, use a non-sensitive prompt to confirm which account or workspace is active and that the expected controls are visible. Do not use sensitive code as a test payload.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Recheck when your setup changes
Repeat the review when you switch plans or accounts, select a new model, add an API key, enable an agent feature, or adopt another IDE integration. Those changes can alter which provider processes the request or which policy applies. Vendor settings and model lists can also change, so follow the current official documentation linked above rather than relying on an old screenshot or a setting name from another product tier.
Quick Recap
Best Value
Rank #4
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
