By Patrick Wahlmueller

AI and Automation in Practice

GitHub Copilot Training Data Policy: What Changes April 24
Technical Insights 3 min read

GitHub Copilot Training Data Policy: What Changes April 24

Starting April 24, 2026, GitHub uses consumer interaction data for AI training. What that means, who is affected, and how to opt out — in case you haven't already.

There is a change in GitHub Copilot data training. Starting April 24, 2026, your prompts, outputs, and code snippets are included in AI training through your interaction with GitHub Copilot.

This applies if you are opted in. If you have already opted out, you remain opted out. Business and Enterprise plans are contractually excluded from training.

Quick context, since panic is spreading in my LinkedIn feed.


The most important points first

In general, all GitHub Copilot users have a switch to allow training or not (Allow GitHub to use my data for AI model training). GitHub explicitly included this link in the announcement.

For individual users, training was already enabled by default. For companies (Business and Enterprise), training of checked-in code was not enabled by default.

EVERYONE can change this setting if they want.

❗ GitHub does not change these settings for you. If you are opted out, you stay opted out. ❗ What changed is WHICH data is used for training.

This data is used only by Microsoft and is not passed on to Anthropic or OpenAI.

From now on, additional data is used for training. You could describe it as Microsoft using GitHub Copilot usage patterns to improve GitHub Copilot.

What is changing specifically

Until now, GitHub used interaction data mainly from Microsoft-internal users for model training. This is now being expanded to the broader user base. Again: only with opt-in; if you already opted out, you remain opted out.

Affected starting April 24, 2026

  • Copilot Free - without opt-out
  • Copilot Pro - without opt-out
  • Copilot Pro+ - without opt-out

Exception: if an individual user is in an enterprise GitHub environment, training is also not enabled.

❗ All Copilot Business and Copilot Enterprise customers have contractual guarantees that data is not used for training.

What data is collected specifically

  • Accepted or modified code suggestions
  • Your prompts to Copilot including code context
  • Cursor position and surrounding code
  • Comments and documentation
  • File names, repository structure, navigation
  • Chat interactions, inline suggestions
  • Thumbs-up/down feedback

See GitHub announcement (EN)

What stays the same

Not everything changes. A few important exceptions:

Not affected

  • Copilot Business and Copilot Enterprise
  • Users who had already opted out before -> preference remains in place, no need to opt out again
  • Content from Issues, Discussions, or private repositories at rest

Opt-out

Ok, this section title was suggested by my LLM of choice 😅. Here is the opt-out page.

My opinion

This topic is heavily amplified, but with an individual account you can switch it off if you want. Companies additionally have contractual guarantees that training is disabled. From what I found via LLM research, competitors handle this similarly.

More information

If you want to understand the process in more detail.

Thanks to Peter Forster


Do you want more?


Sources: GitHub Blog - Updates to GitHub Copilot interaction data usage policy, published by Mario Rodriguez, March 25, 2026.