All AI-generated code should be Open Source

1193 points · 636 comments on HN · read original →

Points and comments are a snapshot, not live.

AI-generated code should be open source because models trained on open-source code without attribution, and AI code may lack copyright.

The author argues AI code should be open source, citing that models were trained on open-source code and Stack Overflow content under Creative Commons licenses requiring attribution, calling proprietary AI code 'license laundering.' He notes that purely AI-generated code may not be copyrightable, citing the Monkey Selfie case and Wikimedia Commons policy treating AI images as public domain. He proposes developers release AI code under open-source licenses or CC0, especially when humans contribute little beyond prompts.

What commenters are saying

Commenters split into camps. Some argue that if AI code isn't copyrightable, open-source licensing is meaningless; others counter that publication itself enables openness, and licenses cover liability. Critics compare human learning from open-source code to AI training, questioning why humans aren't similarly obligated. Several note legal uncertainty: Doe v. GitHub suggests LLM output doesn't inherit open-source licenses, while Thaler v. Perlmutter found 100% AI output uncopyrightable. However, human steering can yield copyrightable AI works, per the US Copyright Office.