GitHub Will Harvest Your Copilot Data to Train AI – What Developers Need to Know

GitHub Copilot data usage policy update

Effective 24 April 2024, GitHub will collect interaction data from Copilot users of the Free, Pro and Pro+ plans to train its AI models.

Scope of affected accounts

Copilot Free, Pro, Pro+ – new policy applies.

Copilot Business, Copilot Enterprise – unchanged, governed by existing contracts.

Students and teachers – exempt.

Data collected

Model outputs that users accept or edit.

Model inputs, including the code snippets shown by Copilot.

Code context surrounding the cursor (e.g., surrounding lines, file content).

Comments, documentation, and other free‑form text written in the repository.

File names, directory structure, and repository metadata.

Interactions with Copilot features such as the chat window, inline suggestions, or command palette.

Explicit feedback signals (thumbs‑up, thumbs‑down, rating, etc.).

Impact on private repositories

Private repositories remain visible only to the owner, explicitly granted collaborators, and, for organization repos, members with appropriate permissions. However, the collected interaction data is extracted from those repositories and sent to Microsoft for model training, regardless of the repository’s private status.

Opt‑out mechanism

Users can disable data collection by navigating to https://github.com/settings/copilot/features, locating the “Privacy” section, and turning off the toggle labeled “Allow GitHub to use my data for AI model training”.

Industry context

Similar data‑usage and opt‑out policies are in place for other AI‑assisted development tools such as Anthropic’s Claude, JetBrains AI, and Microsoft’s own Copilot offerings.