All AI-generated code should be Open Source
TL;DR Highlight
A blog argues AI-generated code should be open source; most HN commenters disagreed.
Who Should Read
Developers wondering about copyright and licensing for internal or commercial code built with AI coding tools
Core Mechanics
- Claim: AI code should be open source
- The author strongly believes all AI-generated code should be open source.
- Developers and companies should release it under open source licenses or CC0.
- Reason one: training data
- Models were trained on open source code scraped without permission or attribution.
- Stack Overflow content requires attribution under Creative Commons, so closing the output feels like license laundering.
- Reason two: pure AI code has no copyright
- Code not written by a human should not be copyrightable.
- The author cites the Monkey Selfie dispute, later cited by the U.S. Copyright Office on AI.
- Wikimedia Commons tags purely AI-generated images as Public Domain.
- Laws differ by country
- The UK grants limited copyright to computer-generated works.
- So explicit release would make things fair across jurisdictions.
- Scope: code with little human effort
- The author admits code is rarely 100% manual or 100% automated.
- Code where humans only wrote prompts, without review or edits, should definitely be open source.
Community Reactions
- Comment quoting Copyright Office guidance
- Purely AI material is not copyrightable, but human expression, selection, arrangement and modification are.
- Prompts alone do not provide sufficient control with current technology.
- Real effort goes into AI-built software
- One commenter spends 20 minutes on agentic coding against hours or days of testing and product thinking.
- They ask whether private finance or school-schedule code must be opened too.
- Legal contradiction argument
- One commenter said licenses depend on copyright, so uncopyrightable AI code cannot be licensed.
- Another noted open source and public domain are different things.
- AGPL author's experience
- A company shipped their AGPL code without attribution, then had an LLM clone it after violation notices.
- The clone was buggier but still felt like their creative work.
- Case law mentioned
- A comment says Doe v. GitHub found LLM output is not automatically open source.
- Thaler v. Perlmutter showed 100% AI output cannot get copyright.
How to Apply
- Audit AI usage in your repos
- Today, pick modules mostly generated by AI and note how much humans reviewed or changed them.
- Use human selection and modification as your yardstick, as in the Copyright Office guidance.
- Record AI involvement
- Note AI generation and human edits in commit messages or the PR template.
- This gives you evidence if copyright or licensing questions arise.
- Decide on publishing with legal
- Before opening or licensing internal code, ask legal whether purely AI code is copyrightable.
- Remember the post is the author's opinion, not a legal obligation.
Terminology
Related Papers
Claude was trained on protein sequences at scale
Next-amino-acid tests on obscure proteins show only the latest Claude models predict them
Remembered as Brief: Stored Preferences SilentlyRemove Decision-Critical Information
A stored 'keep it short' preference silently drops warnings; listing must-know points first restores content.
What thousands of conversations look like inside a language model
Mapping Qwen3-1.7B states showed Chinese sitting apart from English, while German and Croatian overlapped it.
AI, activity, and outcomes in software engineering
AI sped up writing code, but shipping speed did not rise as much, says a talk
LLMs Play Zachtronics' TIS-100
16 LLMs played TIS-100 and set 133 new Pareto frontier points across 18 puzzles
AI usage generates a fragile codebase
Generating all code with AI left shallow tests, complex code, and less human understanding, says one dev.
Accepted but Wrong: Silent Failures of LLM Code Repairs after API Migrations
One release-note edge input cut wrong LLM repairs passing the gate from 0.410 to 0.065
Qwen 3.8 27B on my laptop can interpret model weights, too
Local Qwen 3.8 27B identified 3 of 4 games from raw network weights alone