One team, eight AI training settings

GitHub now trains its AI on Copilot data unless each user opts out. For teams on individual plans, that’s one switch per developer. What a dedicated forge on Servala changes, and what it doesn’t.

By Aarno Aukia • Nov 04, 2026

Your team’s AI training setting is eight settings. Your forge doesn’t have to be.

Since April 24th, 2026, GitHub uses what its Copilot users type and receive to train its AI models, unless each user opts out. That covers Copilot Free, Pro and Pro+, and Copilot Max according to GitHub’s settings documentation. Copilot Business and Copilot Enterprise are not affected.

For one developer, that is a preference: open the settings, decide, done. For a team, it is a different kind of question, and it is the reason a team recently gave VSHN for looking at Codey, the managed Forgejo service you can order on Servala. The question is not whether GitHub’s policy is good or bad. It is who gets to decide.

What GitHub changed

GitHub announced the change on March 25th, 2026 in its blog and in a changelog entry on its Privacy Statement and Terms of Service. Read together with its Copilot settings page, they say:

  • What is used. Interaction data: inputs, outputs, code snippets and the context around them. GitHub’s announcement lists the code around your cursor, comments, file names, repository structure and the feedback you give on suggestions.

  • Who is covered. Users on Copilot Free, Pro and Pro+. The settings page adds Copilot Max.

  • Who is not. Copilot Business and Copilot Enterprise users. The changelog says organization-provided accounts stay under GitHub’s Data Protection Agreement, and the settings page says the training setting is not even displayed for those plans. The announcement also excludes interaction data from enterprise-owned repositories, whoever’s plan is in use.

  • Private repositories. Their content “at rest” is not used. Interaction data generated while someone works in a private repository with Copilot is covered, unless that person opts out or the repository is enterprise-owned.

  • Who receives it. GitHub and its affiliates, which GitHub says include Microsoft, and not third-party model providers.

  • How to opt out. In your personal Copilot settings, set “Allow GitHub to use my data for AI model training” to Disabled. Anyone who had already opted out of data collection for product improvements keeps that choice.

Why one switch per person is a team problem

The setting sits in each person’s account. A team of eight developers on individual plans has eight copies of that setting, each owned by a different person, and the team’s actual policy is whatever each of them last chose. Someone who joins next month brings whatever their own account says: the default, unless they opted out at some point before. The team finds out only by asking.

None of GitHub’s three pages describes a way for an organization to set this for members who are on individual plans. The route they do describe is a Copilot Business or Enterprise seat, where the question does not come up. That is a real answer, and for a team that wants Copilot it may be the right one: it moves the decision from eight people to one contract.

Moving the code somewhere else does not change it. GitHub’s settings page says the Copilot settings “apply wherever you use GitHub Copilot”, so a developer with a personal Copilot plan who edits code hosted on Codey is under the same training setting as on GitHub. Where the repositories live and which AI assistant your developers use are two separate decisions, and a team that cares about the second needs an answer for the AI assistant as well, whatever forge it picks.

What a dedicated forge on Servala changes

What a forge of your own changes is the first decision: who sets the terms for the place your code, issues and pull requests live.

A Codey instance is a dedicated Forgejo instance, not a tenant on a shared platform, and your team has full administrator control over its repositories, users and settings. It runs on Cloudscale, in data centers in Switzerland. VSHN runs the updates, backups, monitoring and security patches, and you can export everything and run Forgejo yourself at any time.

Forgejo itself is free software under GPLv3, governed by its contributors. The non-profit Codeberg e.V. holds its domains and provides resources, and Forgejo’s governance stays independent of it. Every feature is available to every user.

Defaults you can read

The defaults of a Codey instance come from three documented places:

  • The Forgejo project’s releases

  • The platform defaults that Servala manages, as listed in the Codey service documentation: offline mode with no external avatar or CDN requests, restricted new accounts, and a system-managed admin account

  • Your own admin settings

You can read all three, and the last one is yours.

From order to running forge

Codey starts at CHF/EUR 39 per month, billed by the hour, and you order it yourself on Servala.

Moving over is short for a small team. Forgejo’s built-in migration imports repositories, issues, pull requests, labels, milestones, releases and Git LFS objects from GitHub. Forgejo Actions is compatible with GitHub Actions workflows, so pipelines carry over with minimal changes. According to the Codey FAQ, most teams are up and running within a day.

Order your Codey instance on Servala

Sources:

Related Links

Codey

Service

Codey is a secure, fast and reliable code collaboration platform made and hosted in Europe, …

Cloudscale

Provider

cloudscale is designed to be easy to use not just from a technical perspective: By …

Questions about this article?

Have questions or need more information about the topics covered in this article? Get in touch with us!

Ready to Get Started?

Subscribe to our newsletter to stay informed.