February 13, 2026
Good vs Evil vs AI
Ask HN: Is there a no-LLM license yet?
Dev world split: the dream of a “no‑AI” license vs reality
TLDR: People asked if a license can forbid AI training on code; veterans warned non-standard licenses kill adoption and may be ignored. Others cited fair-use rulings and muddy law, sparking a split between idealists and pragmatists. It matters because creators want protection without scaring users—or fighting billion-dollar AI firms.
Can you slap a “no AI training” sticker on your code and call it a day? The thread lit up with devs rehashing the infamous JSON license’s “Good, not Evil” clause—yes, the one that’s so fuzzy it’s basically a compliance nightmare. One commenter even linked the original JSON license for a refresher, and the vibe was: ethics clauses sound noble, but they break in the real world.
The strongest pushback came from the “keep it simple” crowd: non-standard licenses scare companies, period. One pro bluntly said they’ll skip any project that isn’t standard open source, like MIT or Apache. On the flip side, the cynics argued a no-AI training license would be ignored by the tech giants anyway—“try suing,” they snarked, pointing to big publishers’ failed swipes at Meta. Legal realists jumped in too, noting courts have called training fair use when data is legally obtained. Translation: even the law is muddy.
For comic relief, someone dropped a wink to an AGENTS.md doc, and the thread devolved into memes about “Good vs Evil” clauses breaking builds and lawyers noping out. In short: developers are split between protecting their work from AI, keeping adoption high, and not becoming their own legal department. Drama, served hot.
Key Points
- •The post asks about the existence of a “no-LLM training” license and compares it to prior restrictive-use licenses.
- •It cites and quotes the JSON license, which adds a “Good, not Evil” clause to an otherwise permissive license.
- •The post argues subjective restrictions make compliance difficult because they are not objectively measurable.
- •It suggests LLM training may currently be objectively measurable but could become subjective over time.
- •It notes that how copyright law applies to LLM training remains unresolved and historically functions as a powerful constraint.