LLMs Launder Open‑Source Code

Updated: 2026.04.01 19D ago 2 sources
Developers ran an existing LGPL codebase and its tests through a large language model, then published the result as a claimed "ground‑up" rewrite under a permissive license. The move raises an unsettled legal question: can copyrighted source be converted into a new, relicenseable work by processing it with an LLM without clean‑room conditions? — If permitted, the practice would let actors strip value from open‑source projects and relicense or commercialize them, undermining contributor rights and the incentives that sustain the commons.

Sources

AI Can Clone Open-Source Software In Minutes
BeauHD 2026.04.01 90% relevant
The article documents an experiment (malus.sh by Dylan Ayrey and Mike Nolan) that uses AI to recreate open‑source projects into proprietary, attribution‑free variants — a direct instance of the existing idea that large models can 'launder' open‑source code into products that bypass original licensing and attribution.
Python 'Chardet' Package Replaced With LLM-Generated Clone, Re-Licensed
BeauHD 2026.03.06 100% relevant
Release of chardet v7: maintainers claim a 43x speedup and an MIT‑licensed rewrite produced by running the original code and test suite through the Claude LLM.
← Back to All Ideas