Most Open-Source Projects Welcome AI-Assisted Code—But Demand Transparency and Human Review

Generative AI has begun reshaping how developers contribute to open-source projects, and maintainers are scrambling to write rules that balance innovation with quality. A new empirical study examined how the top 1,000 repositories on GitHub are handling the influx of AI-generated code.

What the data shows#

Researchers identified 118 distinct AI policies across the studied projects. The findings paint a picture of cautious optimism: 78% of projects explicitly permit AI-assisted contributions, signaling that the open-source community largely views GenAI as a productivity tool rather than a threat. Only 22% of projects actively discourage or ban AI-assisted work outright.

However, permission comes with strings attached. More than half of the projects—51%—now require contributors to disclose when they've used AI tools. Even more striking, 74% mandate that a human reviewer remain involved in the contribution process, ensuring that code generated by machine models still passes through trained human eyes before merging.

The emerging consensus#

The study suggests that open-source maintainers are converging on a pragmatic approach: welcome AI as a labor-saving device, but enforce transparency and human oversight. Disclosure requirements help reviewers understand the provenance of code changes and apply appropriate scrutiny. The human-in-the-loop mandate reflects concern that fully automated contributions—no matter how syntactically correct—may harbor logic flaws, performance issues, or security vulnerabilities that a machine cannot reliably catch.

Why this matters#

As GenAI tools become cheaper and more accessible, the boundary between human and machine authorship will continue to blur. These emergent guidelines from major projects offer a blueprint for how other teams might govern AI contributions without stifling the efficiency gains that make these tools valuable. The emphasis on disclosure and human judgment suggests the open-source world is learning to trust AI as an assistant, not a replacement for engineering judgment.