GitHub will use more Copilot interaction data for model training by default

GitHub announced on March 25, 2026 that it is changing how GitHub Copilot interaction data is used for model training. Starting April 24, 2026, interaction data from Copilot Free, Pro, and Pro+ users will be used to train and improve GitHub’s AI models unless those users opt out in privacy settings. Copilot Business and Copilot Enterprise users are not affected by this update.

The scope of data GitHub says it may use is broad. The company listed accepted or modified outputs, prompts and code snippets sent to Copilot, code context near the cursor, comments and documentation, file names, repository structure, navigation patterns, interactions with Copilot features such as chat and inline suggestions, and explicit feedback such as thumbs up or thumbs down.

GitHub also tried to draw a line around what is excluded. It said the new program does not use interaction data from Copilot Business, Copilot Enterprise, or enterprise-owned repositories. It also does not use data from users who opt out. At the same time, GitHub explicitly noted that private repository content "at rest" is not used, but code from private repositories can still be processed when Copilot is actively being used, and that interaction data could be used for training unless a user opts out.

Another important detail is data sharing. GitHub said the data used in this program may be shared with affiliates in its corporate family, including Microsoft, but not with third-party AI model providers or other independent service providers. The company argued that using real-world interaction data should improve accuracy, security, and relevance across development workflows.

For developers, the practical impact is straightforward: the default for individual-tier Copilot usage is moving closer to participation in model improvement unless privacy settings are changed. That makes this less of a minor policy cleanup and more of a meaningful platform-level shift in how coding assistant telemetry, workflow context, and edited outputs are recycled into future model behavior.