AI-Generated Code and the Concept of a Work

Generative AI does not create a new problem in software copyright; it makes visible a crack that has been there, and ignored, for forty years. Code is at once a text that is read and a machine that runs; copyright was built to look at only one of those.

Where originality is densest — architecture, functional design — protection does not run, because that is idea and principle. Where protection does run, the source-code text, originality is at its weakest. Good software engineering treats predictability as a virtue; copyright looks for uniqueness.

The United States, the EU and China reach different results, but they ask the same question: does human control reach the expression? The UK’s computer-generated-works provision, after thirty-five years, is now a failed experiment whose own government proposes to repeal it.

Rather than a new regime, the study proposes a graduated-contribution model: originality arises where control reaches expression. Replicability, traceability and the direction of the contribution are the working tests. The full paper is below.