All AI-generated code should be Open Source

Claims that all AI-generated code should be open source have ignited debate over copyright, licensing, and fairness to the open‑source projects used as training data. Commenters argue over whether model output can be copyrighted at all, whether copyleft terms (like the GPL) should “stick” to LLM output, and if companies are effectively laundering restrictive licenses via AI tools. Many see current law as unclear and unevenly enforced, worry about the economic impact on both open‑source maintainers and proprietary vendors, and question whether treating AI training like human learning is a valid legal or ethical analogy.

Legal status of AI‑generated code

  • Strong disagreement on whether AI output is copyrightable.
    • Some cite US Copyright Office guidance: purely AI‑generated works lack copyright; human “steering” and selection/arrangement can be protected, but prompts alone often aren’t.
    • Others note this is still largely untested in courts; current cases (Doe v. GitHub, Thaler/Allen v. Perlmutter, etc.) are limited and don’t settle everything.
  • Several point out: if there is no copyright, open‑source licensing is legally meaningless; such code would effectively be public domain, not “open source.”

Interaction with open‑source and copyleft licenses

  • One camp argues: models trained on GPL/AGPL or CC‑BY‑SA code should inherit those licenses, otherwise it’s “license laundering.”
  • Counterpoint: training on copyrighted code is more like human learning or fair use; courts so far have not treated LLM output as automatically infringing or subject to original OSS licenses.
  • Distinction emphasized between:
    • Public domain (no rights, no obligations)
    • Permissive OSS (MIT/BSD)
    • Copyleft (GPL/AGPL), which relies on enforceable copyright.

Ethical and fairness debates

  • Many feel it’s ethically wrong that FOSS and CC‑licensed work was bulk‑scraped and monetized, then used to replace those same contributors.
  • Others respond that humans have always learned from public code and that tools (LLMs, compilers, scaffolding generators) don’t themselves gain rights; the human user or employer should own the resulting software.

Practicality and enforceability

  • Skepticism that “all AI‑generated code must be open source” could ever be enforced, especially against well‑funded violators.
  • Trade secret law is raised as an alternative: even if no copyright applies, companies can still keep AI‑generated code secret if it’s not distributed.
  • Reverse engineering with LLMs is said to make closed source less protective anyway.

Value and authorship of AI‑assisted work

  • Some say AI‑produced code is low‑value “slop” unless heavily curated; the real value is in design, domain knowledge, and ongoing refinement.
  • Others warn that if AI‑generated artifacts lack copyright, entire AI‑based business models (e.g., AI music catalogs) may be legally fragile.
  • There’s persistent tension between seeing AI as mere tool vs. seeing it as an intermediate “author” that breaks traditional IP assumptions.