Copyright Office says pure AI need not apply

Simply popping a prompt into ChatGPT and then submitting whatever it spits out won’t cut it—that’s true whether one is churning out a college essay or applying for US copyright protections.

A report from the US Copyright Office last week held that works created using AI—no matter how intricate the prompt—cannot be copyrighted without evidence of human creative contributions. The conclusion, an interpretation of existing copyright law, comes after nearly two years of review and thousands of public comments.

The report nodded to the growing use of AI tools in creative industries like film and music, but noted there must be “human control over the expressive elements” for copyright protections to apply.

“Whether human contributions to AI-generated outputs are sufficient to constitute authorship must be analyzed on a case-by-case basis,” the office wrote.

The Copyright Office, a division of the Library of Congress (and thus under control of the legislative branch), has been thrust into a spotlight since the rise of generative AI opened up a mess of questions around staid copyright laws.

But the office hasn’t deviated from its original stance that wholly machine-made art isn’t eligible for copyright. The office also determined in its report that laws already on the books covered enough to answer these questions without new legislation. “Questions of copyrightability and AI can be resolved pursuant to existing law, without the need for legislative change,” the authors of the report wrote.

More than a prompt: The Copyright Office concluded that a prompt alone, no matter how creative, is not enough for consideration, citing comments in support from Universal Music Group and the American Society of Composers, Authors, and Publishers.

“No matter how detailed, text prompts cannot fully define an actual piece of music,” a UMG rep commented.

More to come: This report was part two of the Copyright Office’s findings. The first concerned digital replicas, and called for new federal legislation. The next part will explore the use of copyrighted works in training data for AI.