Clojurians Log v2
ai-skeptics 2026-06-27
The legal aspects of LLM usage is bewildering, and probably contributes to why feet are getting cold when it comes to disclosing AI contributions in code bases. Below are a couple of recent HN discussions that helped me catch up a bit on the copyright side of things in the US, though the same questions exist elsewhere. • Honesty gets Emacs patch rejected: https://news.ycombinator.com/item?id=48681557 • Ask HN: Is "no source code was copied" still a sufficient copyright defense?: https://news.ycombinator.com/item?id=48689939 Again AI blows out of proportion all the warts in our ways of doing things. I'd say most of the world is still in the FA part of FAFO, while the US is edging towards FO, which may lead us into a stricter way of recording AI contributions in order to refine the concept of what can be copyrighted. Not sure what other outcomes could preserve the current state of affairs.
The more I think about this the more I see two kinds of outcomes: 1/ stronger incentives, possibly legal, to distinguish AI code from human code 2/ important changes to the legal frameworks to deal with copyright, IP, and responsibility issues. And that's not an either/or, both will probably occur. For now all the little imperfections in our ways to deal with these issues today are being amplified by AI.
Meanwhile the amount of projects where human intervention is limited to the spec/design/plan phase increases.
Apart from open source, companies are still protected through contract and trade-secret law. So in general business I don't think suffers much from the loss of copyright on code generation. Because of that, I'm not sure it'll change. Generally laws around these things are designed to encourage the creation and innovation of new work through rewarding the creators with a path to monetization. I guess some businesses might need to hand out sources like for SDKs and libraries, or might be more worried about on-prem offerings if their code is public domain.
Hard to see how this will turn out, litigation from those that have something to defend is bound to happen though.
Ya, I'm just trying to think and I feel not many company anymore aren't protecting their code behind closed source APIs
But yes, it's all being figured out to some extent
If you get sued for infringing code, in the US, the person saying you stole their code has to register their code and disclose each part of it that was generated by AI. And the defendant can argue and ask for a panel of expert review to assess if more code was generated by AI and so on. Only after you've shown the code is human written, then you can claim the copyright. So ya, AI generated code in the US is public domain, unless you've heavily modified it in novel ways, so not just fixing some typos, bugs, and style.
I always thought that US was very litigation heavy wrt IP rights, but I haven't heard about big cases against LLM uses, which is surprising