Overview
- GitHub announced it will begin using Copilot interaction data from Free, Pro, and Pro+ accounts to train its AI models by default on April 24, with an option to turn it off in Settings under Privacy.
- Copilot Business and Copilot Enterprise are excluded from the update, and The Register reports that student and teacher accounts are also spared.
- The training data can include accepted or edited suggestions, prompts and shown code snippets, code near the cursor, comments, file names and repository structure, feature use such as chat, and thumbs-up or down feedback.
- GitHub says it will not use content from private repositories when they are idle, yet Copilot processes private code as you work and those live interactions can be used for training unless you opt out.
- GitHub cites better suggestion acceptance after using Microsoft employee interaction data and says user interaction data may be shared with corporate affiliates like Microsoft but not with outside AI model providers.