Starting April 24, GitHub will use interaction data from Copilot Free, Pro, and Pro+ users to train its AI models. That data includes accepted outputs, code snippets, cursor context, file names, repository structure, comments, and thumbs up/down ratings. Copilot Business and Enterprise users are not affected. Users who previously opted out retain that preference automatically.
The policy shift follows internal results GitHub already has in hand: training on Microsoft employee interaction data produced measurable acceptance rate increases across multiple programming languages. The article is worth reading in full because it draws a precise line between what is and is not in scope, specifically the distinction between private repository code 'at rest' versus code actively processed during a Copilot session, which is eligible for training unless you opt out via Settings under Privacy.
Interaction data may be shared with GitHub affiliates including Microsoft, but will not go to third-party AI providers. GitHub says it will also begin incorporating data from its own employees. The opt-out mechanism is live now at github.com/settings/copilot.
[READ ORIGINAL →]