December 28, 2025
Open source vs AI: fight!
Ask HN: Anti-AI Open Source License?
Dev wants “No AI” on their code — open‑source purists clap back
TLDR: A developer asked for a license that allows open source but bans AI training. The community fired back that such restrictions aren’t truly open source, suggested GPL/AGPL as partial defenses, and warned big AI will ignore licenses anyway—highlighting a clash between open‑source ideals and the reality of AI data scraping.
A Hacker News dev asked the internet’s favorite courtroom: can you open‑source code and ban AI from training on it? Cue the purists vs pragmatists cage match. The purists flashed the Open Source Initiative rulebook, shouting that any “no AI” restriction breaks the sacred creed of open source — see the official OSD. One commenter summed it up: if it’s open source with extra strings, it’s not open source. Period.
On the practical side, doom‑and‑gloom realists said even tough licenses like GPL/AGPL (which require sharing changes) won’t stop AI giants from slurping everything anyway. “They’ll scoop it up regardless,” sighed one user. Others floated clever hacks: make AI models trained on the code share back under a compatible license. Then the lawyers crash the party: folks warned fair use (the legal idea that lets you reuse stuff in certain ways) could make any “anti‑AI” clause either redundant or legally void.
The thread devolved into spicy semantics: is “training” the same as “running” software? One commenter invoked Freedom 0 (the Free Software Foundation’s “run for any purpose” principle) while another argued training isn’t running at all. Meanwhile, the memes rolled in: “slap a ‘No AI, No vibes’ sticker” and pray. The vibe? Open‑source values vs AI reality — and nobody’s winning cleanly.
Key Points
- •The post asks if an open-source license can prohibit using code for AI training.
- •Responses state that use-case restrictions would not meet OSI/FSF definitions of open source/free software.
- •Suggestions include AGPL-like conditions requiring AI-trained systems to be licensed compatibly, rather than an outright ban.
- •Legal outcomes on fair use could render AI-specific copyleft clauses redundant or unenforceable.
- •If AI training is not fair use, trained models might be derivative works; enforcement and compatibility remain complex.