RE: https://infosec.exchange/@david_chisnall/117066183348062085

It cannot be stated enough that you must assume a) that your LLM got trained on code that was licensed under licenses like the GPL, and b) that your LLM will reproduce chunks of code that are large enough to be copyrightable from its training set. TL&DR: your LLM project is now covered under the GPL (or worse, AGPL, or any other random license, whether FOSS-approved or not). If you open source your code, your whole codebase might or might not be tainted by that license.

This alone is a very good reason for open source projects to not accept large (larger than 10-15 lines) LLM-coded patches at least until we see a bunch more court cases to settle what this exactly means, unless of course you don't care about licensing details like this. If you use OSS code and plan to redistribute derivative works, know that such a project's license text is not worth much anymore.

What it means for private code you plan to make money off, talk to your lawyer. You're a company, you can afford to.