GitHub is exploring stricter controls for pull requests and introducing AI-powered filters after maintainers reported being overwhelmed by a flood of poor-quality, AI-generated submissions.
GitHub played a significant role in popularizing AI-driven code generation with tools like Copilot. Now, the platform is contemplating measures to restrict, at least temporarily, this very influx of AI-generated content.
GitHub is contemplating a potentially contentious proposal to give repository and project maintainers the ability to either delete pull requests (PRs) or disable PR submissions entirely. This move aims to tackle the growing volume of substandard, often AI-generated, contributions that are proving unmanageable for numerous open-source initiatives.
Last week, GitHub product manager Camilla Moraes initiated a community discussion thread to gather input on proposed solutions for the “rising tide of low-quality contributions.” This increase, she noted, is imposing considerable operational burdens on open-source project and repository maintainers.
“We’ve been hearing your concerns about the significant amount of time spent reviewing contributions that fall short of project quality benchmarks,” Moraes explained. “These issues stem from various factors, including non-adherence to project guidelines, frequent abandonment soon after submission, and a prevalence of AI-generated content.” She then outlined the planned solutions.
AI Challenges the Core Trust in Code Review
Several participants in the discussion thread concurred with Moraes’s assessment that AI-generated code is creating substantial hurdles for maintainers.
Jiaxiao Zhou, a software engineer at Microsoft’s Azure Container Upstream team and a maintainer for Containerd’s Runwasi project and SpinKube, for instance, highlighted that AI-generated code renders the traditional line-by-line review process for shipped code unsustainable for maintainers.
Zhou enumerated several reasons: the erosion of trust in code reviews, where reviewers can no longer assume contributors fully grasp their submissions; the inherent risk of AI-generated pull requests appearing structurally sound but containing logical flaws or security vulnerabilities; and the reality that while mandatory for production code, exhaustive line-by-line reviews do not scale effectively with large, AI-assisted changes.
To address these immediate challenges, Moraes stated that GitHub intends to implement configurable pull request permissions. This will enable more granular control over PR access, allowing maintainers to restrict contributions solely to collaborators or to disable PRs for specific scenarios, such as mirror repositories.
“This enhancement will also eliminate the need for various open-source projects to develop custom automations for managing contributions,” Moraes noted.
Community Expresses Concerns Over Disabling or Deleting Pull Requests
However, the specific proposal to disable PRs encountered skepticism.
A user named ThiefMaster suggested that GitHub should avoid outright restricting access to previously opened PRs, as this could result in content loss or access denial. Instead, the user proposed that GitHub should ensure users can still access them via a direct link.
Moraes appeared receptive to this suggestion, indicating that GitHub might incorporate it into their plans.
Additionally, GitHub is contemplating offering maintainers the ability to directly remove spam or low-quality PRs from the interface, aiming to improve repository organization.
This particular suggestion was met with even stronger apprehension from users.
While ThiefMaster proposed that GitHub could permit maintainers to delete a PR within a limited timeframe, possibly due to inactivity, other users like Tibor Digana, Hayden, and Matthew Gamble expressed outright opposition to the idea.
Moraes’s long-term proposals, which included leveraging AI-based tools to assist maintainers in filtering out “unnecessary” submissions and prioritizing valuable ones, also drew considerable criticism.
While Moraes and GitHub contend that these AI tools would reduce review time, users such as Stephen Rosen argue the opposite. They claim AI tools are susceptible to “hallucinations,” which would still necessitate maintainers reviewing every line of code.
AI’s Role: Minimize Noise, Not Introduce Ambiguity
Paul Chada, co-founder of the agentic AI software startup Doozer AI, asserted that the effectiveness of AI-driven review tools will depend entirely on the robustness of their integrated guardrails and filters.
Without such controls, he cautioned, these systems risk deluging maintainers with submissions that lack project context, consume valuable review time, and dilute truly meaningful signals.
“Maintainers are seeking to reduce cognitive load, not grapple with another system they have to constantly second-guess. AI should function as a sophisticated spam filter or an intelligent assistant, not as a reviewer with ultimate authority. When used judiciously, it filters out extraneous noise; when used carelessly, it introduces new layers of uncertainty rather than resolving existing ones,” Chada remarked.
GitHub has also put forth other long-term suggestions to alleviate the cognitive burden on reviewers. These include enhancing visibility and attribution when AI tools are used throughout the PR lifecycle, and providing more precise controls for defining who can create and review PRs, beyond simply blocking all users or restricting to collaborators only.
