Microsoft's GitHub, the world's largest code repository for software development projects, will start using its users' program codes as training material for its own AI.
The change begins on April 24, 2026, and it does not apply to regular GitHub usage, but only to those users who also use GitHub's own GitHub Copilot.
After the change, all conversations had with Copilot, prompts, and code generated by Copilot will be taken as part of Copilot's AI training material. According to Microsoft's statement, the change will significantly improve the quality of code generated by GitHub Copilot, as it will henceforth be based on real life and conversations developers have already had with the AI.
What makes the change unfortunate is that it is a so-called opt-out solution: all GitHub Copilot users are automatically subject to the new policy, unless they themselves refuse to share their activities as AI training material.
You can opt out of the policy by logging into GitHub and navigating from there to settings and turning the function off (example in the image above).
If you do not opt out of the new operating method, these things will henceforth be used as Copilot's training material:
However, not everything flows to Copilot; for example, private (enterprise) repositories of companies will not in any way become Copilot's training material. Similarly, all information related to business accounts (Business or Enterprise subscription) is excluded from the training material.
In addition, old, private (even for private users) code repositories are excluded from the training material, if Copilot AI is not used with them. Immediately, if Copilot is used in such code repositories, they automatically become training material, unless the user has opted out of using their data as AI training material.
According to the data security statement, data is not shared with parties outside Microsoft: meaning, for example, OpenAI does not get access to the training data, but all internal Microsoft units can utilize the training material collected for GitHub Copilot.
After the change, all conversations had with Copilot, prompts, and code generated by Copilot will be taken as part of Copilot's AI training material. According to Microsoft's statement, the change will significantly improve the quality of code generated by GitHub Copilot, as it will henceforth be based on real life and conversations developers have already had with the AI.
What makes the change unfortunate is that it is a so-called opt-out solution: all GitHub Copilot users are automatically subject to the new policy, unless they themselves refuse to share their activities as AI training material.
You can opt out of the policy by logging into GitHub and navigating from there to settings and turning the function off (example in the image above).
If you do not opt out of the new operating method, these things will henceforth be used as Copilot's training material:
- AI-generated codes that the user has accepted or modified
- all prompts sent to the AI
- all context sent with the prompts, i.e., usually parts of existing code
- information about what code was at the cursor's position in the visible editing window when Copilot offered a code completion
- comments and documentation files added by the user to the code
- file names, repository structure, and information on how the user navigates between different files
- user feedback on Copilot's suggestions (thumbs up and down)
However, not everything flows to Copilot; for example, private (enterprise) repositories of companies will not in any way become Copilot's training material. Similarly, all information related to business accounts (Business or Enterprise subscription) is excluded from the training material.
In addition, old, private (even for private users) code repositories are excluded from the training material, if Copilot AI is not used with them. Immediately, if Copilot is used in such code repositories, they automatically become training material, unless the user has opted out of using their data as AI training material.
According to the data security statement, data is not shared with parties outside Microsoft: meaning, for example, OpenAI does not get access to the training data, but all internal Microsoft units can utilize the training material collected for GitHub Copilot.
