Did you know: GitHub can train AI models on personal Copilot prompts and code by default

If you use GitHub Copilot on a personal plan, GitHub has been able to train AI models on your Copilot interactions since April 24, 2026, unless you switched that off. GitHub’s March 25, 2026 announcement by Chief Product Officer Mario Rodriguez says interaction data, “specifically inputs, outputs, code snippets, and associated context”, from Copilot Free, Pro, and Pro+ users “will be used to train and improve our AI models unless they opt out.” The current Copilot settings documentation lists Copilot Max alongside those three plans. Copilot Business and Copilot Enterprise are not affected.

How to opt out

1. On GitHub, click your profile picture in the upper-right corner, then click Copilot settings (the page is github.com/settings/copilot).
2. Under Privacy, open the “Allow GitHub to use my data for AI model training” dropdown and choose Disabled.

The docs say these settings apply wherever you use Copilot, not per editor. If the option is missing, GitHub says to check that you are not signed in with a Copilot Business or Enterprise license: the setting is not shown for those plans, whose data falls under GitHub’s Data Protection Agreement.

If you had already disabled the older setting for collecting prompts and suggestions for product improvements, the FAQ says that choice carried over and your data is not used for training unless you opt in.

Timing matters. Section J.3 of the Terms of Service says that after you opt out, GitHub will not collect or use your inputs and outputs for training “from the effective date of your opt-out going forward.” None of the pages we read describes removing data collected before that. The Privacy Statement lists a separate right to request deletion.

What counts as interaction data

The blog lists outputs you accept or modify, inputs including code snippets shown to the model, code around your cursor, comments and documentation you write, file names, repository structure and navigation patterns, your use of features such as chat and inline suggestions, and thumbs up or down feedback. The FAQ adds content and context supplied through connected tools and services.

Private repositories and who is excluded

GitHub says it does not train on private repository content “at rest.” The blog chooses that phrase deliberately: code from a private repository that you send to Copilot while working is interaction data and could be used for training unless you opt out.

Excluded, per the blog and FAQ: Business and Enterprise data, enterprise-owned repositories, students on Copilot Student, and teachers who get Copilot Pro free. The FAQ says GitHub does not train on content from any paid organization’s repositories, and that if your account is a member or outside collaborator of a paid organization, your interaction data is excluded even on a Free, Pro, or Pro+ plan.

Where the data goes

The blog says the data may be shared with GitHub affiliates, which include Microsoft, and will not be shared with third-party AI model providers. The FAQ adds that GitHub and Microsoft staff working on model development may access it, and that GitHub may use contracted service providers to help with training on its behalf. The changelog says opt-out preferences travel with data shared with affiliates.

GitHub says it filters API keys, passwords, tokens and personal information, limits and audits access, and de-identifies or aggregates data when possible. Those are GitHub’s descriptions of its safeguards; the pages give no independent audit. None of the pages we read sets a retention period specific to training data.

Fine print worth knowing

Section J.3 also says the opt-out “applies solely” to inputs and outputs and does not cover the content license in Sections D.4 to D.8 or content that is not input or output, “including Content stored in public repositories.” For users in the EEA and UK, the updated Privacy Statement names AI development as a legitimate interest, and you keep the right to object.

GitHub frames the change as matching “established industry practices” and says training on Microsoft employees’ interaction data raised suggestion acceptance rates in several languages. That is GitHub’s claim; the post gives no figures.

Practical takeaway

On Free, Pro, Pro+ or Max and not comfortable with this? Open Copilot settings and set the training option to Disabled now. It has applied since April 24, 2026 unless you acted earlier, and the opt-out works from the moment you set it. Using Copilot through a paid organization? GitHub says your data is already excluded.