OSS projects banning AI-generated contributions
Open source maintainers are increasingly adding policies that reject pull requests generated with large language models, citing floods of low‑quality “slop” contributions, reviewer burnout, and difficulty preserving project design and direction. Critics argue this is unenforceable, exclusionary, and risks driving away both genuine contributors and potential security improvements, especially as attackers and other projects lean on AI for rapid development and bug-finding. Many participants distinguish between banning external AI-written code and using AI privately for reviews, security scans, or maintainer-authored changes, reflecting an unsettled attempt to adapt contribution norms to widespread AI tooling.
Motivations for AI Contribution Bans
- Maintainers report being flooded with low-effort, AI-generated “slop” PRs and spammy offers (“AI agents” generating sites, suggestions, etc.).
- Reviewing now takes far more effort: AI-written PRs can look polished, pass tests, and still be subtly wrong or misaligned with project design.
- Traditional signals of contributor effort and understanding have been “sybil-attacked” by LLMs: garbage and genuinely thoughtful work look similar at first glance.
- High mentoring effort for new contributors feels wasted when the real “author” is an LLM rather than a human learning.
Perceived Benefits of Bans
- Acts as “proof-of-work”: if maintainers are to invest serious review time, they want evidence the contributor did real thinking.
- Reduces burnout and preserves project direction and coherence, especially for small volunteer teams.
- Some see bans as a positive quality signal: projects that disallow AI PRs may be safer to depend on or study.
Critiques of Bans
- Risks driving away sincere contributors and shrinking communities, especially in underfunded projects (e.g., Servo, Redox OS).
- Bans are coarse-grained: they block both low-quality and potentially excellent AI-assisted contributions.
- Some view “hand-written only” policies as Luddite, unserious, or even negligent for security-critical software.
- Enforcement is seen by some as impossible and likely to push people to lie about AI usage.
Nuanced / Conditional Approaches
- Several projects distinguish between:
- External AI-generated PRs (often banned or “restricted”).
- Internal use of AI by trusted maintainers (often allowed or common).
- Suggested best practices:
- Use AI for review, bug-finding, design help, not as primary code author.
- Require issues first, limit PR scope, or restrict who can submit AI-assisted code.
Security, Ethics, and Ecosystem Effects
- Some argue not using AI for automated bug-finding and vuln scanning will leave projects vulnerable, especially as attackers use AI at scale.
- Others worry about unethical training on copyleft/proprietary code and dependence on proprietary “black box” tools.
- Predictions range from:
- More forks that embrace AI and move faster.
- To a future where maintainers mostly run AI themselves, and external “contributions” become issues, prompts, or funding rather than code.