GitHub Copilot Is Training on Your Code: How to Opt Out Before April 24

On March 25, 2026, GitHub announced a significant update to its privacy policy: starting April 24, 2026, interaction data from Copilot Free, Pro, and Pro+ users will be used to train and improve GitHub's AI models.

The key detail that caught the developer community off guard: this is opt-out, not opt-in. Unless you actively disable the setting, your Copilot interaction data will be collected for AI model training.

GitHub says the change is based on positive results from using Microsoft employee interaction data to improve Copilot models. They now want to extend this to their broader user base.

The announcement came through a blog post and a community FAQ discussion — not through email notifications to affected users, which has been a significant point of criticism.

Affected:
- Copilot Free users
- Copilot Pro users ($10/month)
- Copilot Pro+ users ($39/month)

NOT affected:
- Copilot Business users — enterprise data policies remain unchanged
- Copilot Enterprise users — same, enterprise protections apply
- Students and teachers accessing Copilot through GitHub Education — explicitly excluded
- Users who previously opted out of data collection for product improvements — your preference carries over

The distinction matters: if you are paying for Copilot Pro or Pro+, your interaction data is still subject to training collection by default. The fact that paying customers are included alongside free users has been a major point of contention in the developer community.

According to GitHub's updated privacy statement, the following interaction data may be collected when you use Copilot:

Collected during active Copilot sessions:
- Code snippets — the code context sent to Copilot for suggestions
- Prompts and inputs — what you type in Copilot Chat
- Outputs and suggestions — what Copilot generates for you
- File names and paths — the files you are working in
- Repository structure — directory layout and file organization
- Navigation patterns — how you move between files during a session
- Feedback signals — whether you accept, reject, or modify suggestions

Important clarification from GitHub:
- Private repository code "at rest" (stored code you are not actively using with Copilot) is not collected
- However, code from private repos during active Copilot sessions — meaning the context window Copilot reads to generate suggestions — is included in the interaction data

This is a critical distinction. If you have API keys, database credentials, or proprietary logic in files where Copilot is active, that context may be included in training data. Copilot currently has no mechanism to exclude specific files from data collection — it is all or nothing.

Opting out takes less than 30 seconds. Here is the exact process:

Step 1: Log into your GitHub account

Step 2: Go to Settings (click your profile picture in the top-right corner → Settings)

Step 3: Click Copilot in the left sidebar (or go directly to github.com/settings/copilot)

Step 4: Scroll down to the Privacy section

Step 5: Find the toggle labeled "Allow GitHub to use my data for AI model training"

Step 6: Disable the toggle (turn it off)

Step 7: Confirm the change

That is it. GitHub says the change takes effect immediately and they will not use your future interaction data for training.

What about data already collected?
GitHub has not been explicit about whether previously collected interaction data (before you opt out) will be removed from training datasets. The FAQ states that opting out prevents "future" data collection, which implies historical data may still be used. If this concerns you, consider reaching out to GitHub support to request data deletion under your applicable privacy rights.

The community response has been overwhelmingly negative. On GitHub's own FAQ discussion thread, the announcement received significant pushback.

Common complaints include:

• Opt-out is the wrong default. Many developers argue that training on user data should require explicit consent, especially for paying customers. As one Hacker News commenter put it: "I'm paying $10/month and now my code is training their models by default?"

• No email notification. GitHub did not email affected users about the change. Many developers only learned about it through Reddit, Twitter/X, and LinkedIn posts — often from other developers warning their peers.

• Timing concerns. Announcing on March 25 with an April 24 effective date gives users exactly 30 days to discover and act on the change. Given that many developers do not regularly check GitHub's blog, the concern is that millions will be opted in without knowing.

• Private repo implications. The fact that Copilot session data from private repos is included — even though stored code is not — has raised concerns about intellectual property and trade secrets.

• Trust erosion. Several prominent developers and tech commentators have noted that this follows a pattern of eroding trust in GitHub since the Microsoft acquisition, from the original Copilot training controversy to this latest policy change.

This is not GitHub's first controversy over AI training data. In 2022, when Copilot launched, it faced a class-action lawsuit alleging it was trained on open-source code without proper attribution or license compliance. That case is still ongoing.

The current change is different — it concerns user interaction data, not stored repository code. But the sentiment is similar: developers feel their work is being used to improve a product they may or may not use, without meaningful consent.

GitHub's position is that this is "standard practice across the industry" and that the data helps them build better tools for developers. They point to improved model performance after training on Microsoft employee data as evidence.

Critics counter that "everyone does it" does not make it right, and that opt-in consent should be the standard for AI training data collection.

If GitHub's data policy concerns you, here are some alternatives worth exploring:

AI Coding Assistants with Stronger Privacy:
- Cursor — uses its own AI models; offers local-first processing options
- Cody by Sourcegraph — enterprise-grade with explicit data handling policies
- Continue.dev — open-source AI coding assistant that can run fully locally
- Tabby — self-hosted AI coding assistant; your data stays on your infrastructure

For Sensitive Projects:
- Use Copilot with the training data toggle disabled
- Consider self-hosted AI solutions for proprietary codebases
- Review your organization's security policies regarding AI coding tools

The choice depends on your specific needs. For many developers, Copilot with the training toggle disabled is a reasonable compromise. For organizations handling sensitive code, self-hosted solutions may be worth the additional setup.

The GitHub Copilot situation highlights a broader truth: any data you send to a cloud service could be used in ways you did not anticipate. This applies to AI coding assistants, online code formatters, JSON tools, and any web-based developer utility.

This is exactly why we built DevPik's tools to work differently. Every tool on our platform — including our JSON Formatter, JSON Minifier, JSON Compare, and Code Share — runs 100% client-side in your browser. Your code and data never leave your device.

No server processing. No data collection. No training. Just tools that work.

At DevPik, all our tools run 100% client-side — your code never leaves your browser. Try our JSON tools, Code Share, and 28+ other free tools.