Stallman bars LLM-generated code from NonGNU ELPA

Richard Stallman has said NonGNU ELPA will not accept packages developed with large language model assistance, turning aside an inquiry about Clutch, an interactive database client for Emacs. Maintainers who actually run the archive immediately rejected his claim that NonGNU packages must follow the same legal rules as GNU software.

Lucius Chen, Clutch's maintainer, asked whether a GPL-3.0-or-later package whose history includes code from OpenAI Codex and Claude Code, disclosed in Assisted-by headers, could be included. Clutch offers SQL editing buffers, an interactive result grid, schema completion, and staged row edits, with a SQLite path that uses Emacs's built-in support and optional backends for other databases.

Stallman answered that current policy is not to include LLM-generated code. NonGNU ELPA packages are not strictly part of Emacs, he wrote, but they are closely associated with it, so the project wants them under the same legal policies as GNU packages.

Stefan Monnier strongly disagreed. "The whole point of NonGNU ELPA is that it does not follow the same legal policies," he wrote. Philip Kaludercic seconded that view. Kaludercic had already noted that LLM-generated code cannot satisfy the copyright assignment required for GNU ELPA, so NonGNU would be the only possible home if such work were allowed at all. He also asked whether Chen could understand and maintain the code without ongoing model help.

A non-maintainer separately objected to GitHub hosting and to hundreds of commits attributed to Claude models, citing free-software philosophy and unsettled copyright. Michael Albinus called the GitHub stance impractical given how many packages already live there, while treating the LLM copyright question as a real issue under present criteria.

The result is a split: Stallman states a ban, while NonGNU ELPA's active maintainers deny that GNU assignment rules, or a blanket LLM exclusion derived from them, automatically apply.