GitHub updates interaction data policy for Free and Pro users. Step-by-step guide to opting out of model training and understanding the privacy impact.
What the interaction data policy covers
GitHub has updated how interaction data is handled for Free and Pro Copilot users. Interaction data is the material your sessions produce while you work with the assistant: prompts you type, code context that gets sent with a request, and the suggestions you accept, reject, or edit. That material can inform product quality work and, depending on your settings, can also be used to improve the models that power suggestions.
The distinction that matters is purpose. Diagnostics and product analytics are not the same as training use. A policy update for Free and Pro plans usually clarifies which of those uses apply by default, which require an explicit choice, and where account-level controls sit. Free and Pro users should assume defaults lean toward broader data use unless they change them; paid enterprise-style plans often ship with stricter defaults, but that does not remove the need to verify your own account.
How to opt out of model training
Treat opt-out as a settings task, not a one-time banner click. Sign in to the account that holds your Copilot subscription, open account or Copilot preferences, and look for privacy, data, or model-training controls. Disable any option that allows your interactions, code, or feedback to be used for model improvement. Save the change, then reopen the same page to confirm the toggle still reflects your choice.
- Check both personal account settings and organization or workspace settings if you use Copilot under a team.
- Review IDE and editor extensions: some clients surface a separate privacy or telemetry switch that does not fully mirror the web account control.
- If you use more than one GitHub identity (work and personal), repeat the check on each identity that has Copilot enabled.
- After a major product or policy notice, re-check the same screens; defaults and labels can change when plans or features ship.
Opting out of training does not usually turn Copilot off. It limits whether your prompts and related context may be used to improve future models. Completions, chat, and other features should continue subject to the product’s normal terms and rate limits for your plan.
Privacy impact in day-to-day work
Even with training disabled, prompts still leave your machine for the service to generate a response. That means secrets, proprietary algorithms, customer data, and internal URLs can still be exposed in transit and processing if you paste them into a prompt or if the tool attaches surrounding file context. Opting out of training reduces secondary use of that material for model improvement; it does not make careless prompting safe.
Practical risk reduction is mostly workflow, not policy text. Strip secrets before you ask for help. Prefer small, self-contained snippets over whole files when the problem is local. Keep production credentials and private keys out of editors that auto-attach context. For repositories with strict compliance needs, align Copilot use with your org’s data-classification rules and disable features that send broader codebase context if those controls exist.
A simple review habit for Free and Pro users
Policy updates are easy to miss until a default no longer matches what you assumed. After you opt out, document the date and the setting you changed in your own notes. On a recurring basis—after plan changes, extension updates, or new Copilot features—return to the same privacy screens and confirm training use is still off. If you collaborate under an organization, ask whether org-level policies override personal toggles so you know which control actually wins.
The useful outcome is narrow and concrete: you understand what interaction data includes, you have turned off model-training use where the product allows it, and you still treat every prompt as data that may leave your machine. That combination—account-level opt-out plus careful prompting—is the practical privacy posture Free and Pro developers can maintain without waiting for another announcement.