Re: GCC AI Policy Announcement

On 01/08/2026 22:06, Thomas Koenig via Gcc wrote:
> That is one of the clearer points right now - LLM-generated code cannot
> be copyrighted.

I'm not a lawyer, so take this for what it's worth, not as advice...

I don't think it's a clear cut as that.  Consider the analogies.  A compiler (eg, gcc) compiles a piece of source code to produce an executable.  That source code has copyright owned by its author; the resulting binary also has copyright owned by the author, but gcc (to the extent that has translated the code from one form, source, to another, executable) does not gain a copyright interest in the output - this is, of course, setting aside any library code that might be combined with the compiled code.

Setting aside, for the moment, the training data used, an LLM is arguably similar, it's a tool and as a tool cannot gain any copyright interest, since copyright is held to be an expression of something by a human; but there's still copyright in the result, as a consequence of the input that was used to drive the LLM - how much will depend on what guidance the LLM was given in that process.

R.