GitHub Copilot Data Training Policy: Settings Individual Developers Should Check¶

For / Key Points

For: Individual developers using GitHub Copilot Free, Pro, or Pro+

Key Points:

Since April 24, 2026, covered interaction data is used for AI model improvement by default
Business / Enterprise and paid organization repository data are excluded
Set Allow GitHub to use my data for AI model training to Disabled

On April 24, 2026, GitHub Copilot's interaction data policy change took effect. It applies to individual Copilot Free, Pro, and Pro+ users. If you do nothing, some of your Copilot interactions may be used to train and improve AI models¹.

The practical question is narrow. Which setting should an individual developer check, and when should Copilot be avoided for sensitive code?

Already in effect Treat this as an active setting, not a future announcement.
Affected plans Separate Free / Pro / Pro+ from Business / Enterprise.
Private repo handling Distinguish repository content at rest from snippets sent to Copilot.
Settings check Disable model training use in Copilot settings.

What Changed on April 24, 2026¶

The change is that Copilot inputs, outputs, code snippets, and associated context may now be used for model training and improvement by default for covered individual plans. GitHub announced the policy on March 25, 2026, with an April 24 effective date¹.

This is not a switch for whether Copilot itself works. It is a switch for whether your interaction data can be used to improve AI models. That means the first place to check is not your IDE extension, but your GitHub account's Copilot settings.

The term interaction data is broader than a usage counter. It can include the surrounding code context sent to Copilot and the outputs you accept or modify. The right mental model is: when Copilot needs context to help you, some active work context may be sent to the service.

Affected and Unaffected Plans¶

The policy is easiest to reason about when you separate individual plans from organization-managed plans. GitHub Docs says Copilot Business and Copilot Enterprise customer data is not used to train AI models under the Data Protection Agreement².

Usage type	Training use status	What to check
Copilot Free	Covered	Can be disabled in personal settings
Copilot Pro	Covered	Can be disabled in personal settings
Copilot Pro+	Covered	Can be disabled in personal settings
Copilot Business / Enterprise	Not covered	The setting may not appear for these seats
Free Pro for students and teachers	Not covered	GitHub FAQ describes these users as excluded

Developers who pay for an individual Copilot Pro subscription should check this setting first. Developers only testing the free plan should still check it if they use private repositories or unpublished side projects.

What Counts as Interaction Data¶

GitHub lists multiple kinds of Copilot use data that may be collected when model training is enabled for an individual user¹. This is not limited to prompts or final code.

Inputs sent to Copilot: prompts, code snippets, and context shown to the model
Copilot outputs: accepted suggestions, modified suggestions, chat, and inline completions
Work context: cursor-adjacent code, comments, documentation, file names, and repository structure
Usage signals: feature interactions and feedback on suggestions

GitHub also says inputs and outputs are not shared with third-party AI model providers for their independent training³. At the same time, the policy allows use within GitHub affiliates, primarily including Microsoft. For individual developers, the relevant question is not only whether data is sold externally, but whether it is used by GitHub or Microsoft for model improvement.

Settings Individual Developers Should Check¶

The setting lives on GitHub's website. GitHub Docs instructs individual subscribers to open Copilot settings and set Allow GitHub to use my data for AI model training to Disabled².

Sign in to GitHub
Open github.com/settings/copilot
Find the Privacy or model training section in Copilot settings
Set Allow GitHub to use my data for AI model training to Disabled
Reload the page and confirm the setting remains Disabled

If you previously disabled prompt and suggestion collection, GitHub FAQ says that preference carries over⁴. Still, UI labels and placement can change. If the setting is missing, first verify that you are not signed in under a Copilot Business or Enterprise seat.

Private Repositories and Work Code¶

GitHub is not saying that entire private repositories are pulled from storage for training. It says private repository content at rest is not used, while code snippets sent during active Copilot use may be collected and used if model training is enabled⁴.

That distinction matters in day-to-day work. A private repository is not automatically safe simply because it is private. The question is whether a snippet reached Copilot during an active session. For secrets, unpublished algorithms, or customer-specific logic, disabling model training before using Copilot is the practical baseline.

There is one important correction. GitHub FAQ says it does not train on the contents of paid organization repositories, even if a user works there with a Copilot Free, Pro, or Pro+ subscription⁴. Do not mix up an employer-owned paid organization repository with a private repository owned by your personal account.

The Controversy and the Practical Decision¶

The controversy is mainly about the opt-out design. GitHub provided a control, but the setting is enabled unless affected users disable it. The Community FAQ collects questions about private repositories, employer code, and consent mechanics⁴.

For individual developers, the decision is clearer if you classify your work.

Work type	Practical default	Why
Public OSS or learning code	Optional	Risk is lower when publication is expected
Personal private repos or unpublished products	Disable	Snippets sent to Copilot may be covered
Employer or customer code	Follow organization policy	Confirm paid organization exclusion and internal rules

This setting is separate from whether Copilot is useful. Using AI assistance and contributing active work data to model improvement are two different decisions.

Summary¶

Since April 24, 2026, individual Copilot Free / Pro / Pro+ users should verify their model training setting at least once. If you use Copilot in private repositories, confirm that Allow GitHub to use my data for AI model training is Disabled.

The broader lesson is that Copilot risk management is no longer a one-time tool adoption decision. As AI agents expand from IDE completions to CLI, issues, and pull requests, settings will multiply by feature surface. Development environment operations now need to include which data-use controls are audited, and where those controls live.