GitHub announced on April 2, 2026, that individual Copilot subscribers will have their interaction data collected and used to train GitHub’s AI models starting April 24, 2026. Chief Product Officer Mario Rodriguez published the update in a post on the GitHub Blog. The change covers users on Free, Pro, and Pro+ plans; Copilot Business and Enterprise customers are excluded.
- Starting April 24, 2026, GitHub will collect interaction data from Copilot Free, Pro, and Pro+ users to train its AI models.
- Copilot Business and Enterprise customers are explicitly excluded from this policy change.
- Users who previously opted out of data collection for product improvements will have those preferences automatically preserved without needing to act.
- Collected data will be shared with Microsoft affiliates but will not be provided to third-party AI model providers or independent service providers.
What Happened
Starting April 24, 2026, GitHub will collect and use behavioral interaction data from individual Copilot users on its Free, Pro, and Pro+ subscription tiers to train the AI models that power Copilot’s code suggestions, according to a policy update published by Chief Product Officer Mario Rodriguez on the GitHub Blog.
The announcement extends a training approach GitHub had already been applying with a more limited internal dataset. Rodriguez stated that GitHub spent the past year using interaction data from Microsoft employees before expanding the practice. The updated policy brings individual Copilot subscribers into the same training pipeline for the first time.
Why It Matters
The policy formalizes data collection practices that had not previously applied to GitHub’s individual paying users, aligning Copilot’s individual-tier strategy with behavioral feedback approaches that GitHub had already been applying internally using Microsoft employee data.
Rodriguez cited direct evidence for the value of this approach: “This past year, we’ve started incorporating interaction data from Microsoft employees and have seen meaningful improvements, including increased acceptance rates in multiple languages.” This describes what was observed in practice with internal data, not the result of a controlled study.
The practical effect is that user decisions to accept, reject, or modify Copilot suggestions—along with the surrounding code context—will feed directly into future model versions targeting improved suggestion accuracy.
Technical Details
GitHub specified seven categories of interaction data it will collect under the updated policy: outputs that users accept or modify, inputs sent to Copilot that include code snippets, code context surrounding cursor positions, comments and inline documentation, file names and repository structure, navigation patterns within the editor environment, and explicit user feedback on individual suggestions.
The policy distinguishes between data at rest and data processed during active use. Content from private repositories, issues, and discussions stored at rest will not be collected for training. However, GitHub stated that code from private repositories that Copilot processes during an active coding session is classified as interaction data and “could be used for model training unless you opt out.”
Collected data will be shared with Microsoft-affiliated entities. GitHub stated explicitly that it “will not be shared with third-party AI model providers or other independent service providers.” GitHub also disclosed it will add interaction data from its own employees to the training pipeline alongside the existing Microsoft employee dataset, expanding the internal baseline used for model development.
Who’s Affected
Copilot Free, Pro, and Pro+ subscribers are directly subject to this policy change effective April 24, 2026, while Copilot Business and Enterprise customers, along with content stored in enterprise-owned repositories, are explicitly excluded from the scope of the updated data usage policy.
Individual developers working in private repositories should note the distinction between content at rest and code actively processed by Copilot. Live coding sessions fall within the collection scope under the new policy, making the opt-out option directly relevant for developers working with proprietary or confidential code.
What’s Next
The updated policy takes effect on April 24, 2026, giving users who have not previously opted out of data collection for product improvements until that date to update their preferences through GitHub’s privacy settings before their data becomes eligible for model training.
Users who previously opted out do not need to take action; Rodriguez confirmed those preferences will be honored automatically. GitHub did not disclose a timeline for when model improvements informed by the expanded dataset would be deployed, nor did Rodriguez provide quantified performance benchmarks beyond the reference to improved acceptance rates observed with Microsoft employee data.
