Only with opt-out: GitHub will train Copilot models with user data in the future
Prompts, code suggestions, comments: GitHub will collect data from AI interactions for its own model training in the future. Objection is possible via opt-out.
(Image: Sundry Photography/Shutterstock.com)
GitHub is changing its terms of use and will use user interactions with Copilot for its AI model training starting April 24, 2026. Those who do not want this can object to it now via opt-out.
The change affects the Free, Pro, and Pro+ plans, but not Business and Enterprise. Students and teachers who use the Pro plan for free, as well as users who have already objected to public code matching, are also not impacted.
GitHub states that data collection serves to better understand development workflows and to provide better and safer code examples. This is intended to reduce bugs reaching production. According to the announcement, corresponding tests within the company have proven the desired model improvements.
GitHub also points to similar conditions at parent company Microsoft, at Anthropic, and at JetBrains.
The opt-out can be found in the settings under "Copilot/Features"
Specifically, GitHub uses the following data:
- Private repositories during use
- Accepted or modified Copilot suggestions
- Inputs sent to Copilot, including code snippets
- Code context around the cursor position
- Comments and documentation
- Filenames and repository structures
- Navigation patterns
- Interactions with Copilot features such as chat or inline suggestions
- Feedback on suggestions (thumbs up/down)
Excluded are:
- Content from issues and discussions
- Private repositories when not in use (“at rest”)
- Business and Enterprise plans
- Students and teachers who use the Pro plan for free
The objection can be made via the settings page in GitHub under “Copilot/Features”.
Videos by heise
Security and data protection
Microsoft intends to use the collected data within its own corporate group but not to pass it on to external model operators or hosts. Access to the data is limited to “authorized personnel working directly on model improvement and safety” or for logging and auditing purposes. According to GitHub, there are several layers of security, including filtering API keys, passwords, tokens, and personal data.
In the discussion in the associated FAQs, users criticize in particular that the opt-out cannot be limited to specific repositories but only allows a global yes or no per account. Others also demand, “At least offer a discount if we agree that the models you sell benefit from it.”
(who)
