Judging the Contributor: How Open-Source Communities Govern Access When AI Makes Contributions Cheap
Prof. Michail Batikas
Assistant Professor
Operations, Technology and Innovation Management
Nova School of Business and Economics
Generative AI has made it cheap to produce contributions that look complete. We study how maintainers of established GitHub projects decide whose proposals to accept. We combine a census of who closed 2.3 million unmerged pull requests in 91,970 repositories with the full histories of 366,656 projects active before AI tools. From mid-2025, maintainers refuse more outside proposals within the same project and month, and the rise is not explained by volume or review load. Most of the extra refusals come with a reason. Closers increasingly cite contribution rules and AI policies when they refuse newcomers who arrive without prior engagement, while newcomers who first opened an issue in the project are largely spared. We interpret the pattern through control theory, as tighter output control combined with process control at the project’s front door.
