Codeberg published a guide on shielding free and open source projects from large language model training runs. The post appeared on Hacker News where it received 12 points and one comment.
What the Proposal Covers
The article recommends combining license language with technical signals that explicitly forbid automated scraping for model training. It targets the growing practice of ingesting public repositories without consent.
Maintainers are advised to add clauses that treat model training as a distinct use case separate from normal open source reuse.
License Options Compared
Standard permissive licenses such as MIT and Apache 2.0 contain no restrictions on training. Newer variants add explicit prohibitions.
| License | Training Allowed | Enforcement Path | Adoption Level |
|---|---|---|---|
| MIT | Yes | None | Very high |
| Apache 2.0 | Yes | None | High |
| Custom No-AI clause | No | Legal notice | Low |
| Codeberg-style terms | No | Takedown + notice | Emerging |
Projects using the stricter terms must accept reduced visibility on platforms that automatically filter restricted content.
Technical Signals Recommended
The post lists three concrete steps: robots.txt entries blocking known crawler user agents, repository metadata files declaring training restrictions, and watermark-style comments in source files.
These signals do not replace licenses but give automated systems clearer instructions before ingestion begins.
Trade-offs for Maintainers
Adopting restrictions can slow contributor growth. Some platforms already deprioritize repositories that block training crawlers.
On the other hand, the approach gives authors a documented position if their code later appears in commercial model weights.
Who Should Apply These Measures
Small teams maintaining core infrastructure libraries benefit most when they want to limit commercial reuse without switching to non-open licenses.
Large projects with corporate backing or those already under copyleft licenses gain little additional protection and may lose discoverability.
Practical Next Steps
Review the exact wording suggested in the Codeberg post. Add the license file first, then the technical markers. Test that major crawlers respect the new signals within 30 days.
Monitor download logs for sudden drops that may indicate crawler avoidance.
Bottom line: The approach gives maintainers explicit control at the cost of narrower distribution.
Codeberg’s recommendations remain one of the few public, actionable frameworks aimed specifically at the FLOSS-LLM tension.
Top comments (0)