Generative artificial intelligence output based purely on text prompts — even detailed ones — isn’t protected by current copyright law, according to the US Copyright Office.

The department issued this guidance in a broad report on policy issues regarding AI, focused on the copyrightability of various AI outputs. The document concludes that while generative AI may be a new technology, existing copyright principles can apply without changes to the law — and these principles offer limited protection for many kinds of work.

The new guidelines say that AI prompts currently don’t offer enough control to “make users of an AI system the authors of the output.” (AI systems themselves can’t hold copyrights.) That stands true whether the prompt is extremely simple or involves long strings of text and multiple iterations. “No matter how many times a prompt is revised and resubmitted, the final output reflects the user’s acceptance of the AI system’s interpretation, rather than authorship of the expression it contains,” the report says.

This decision would seemingly rule out protections for works like “Théâtre D’opéra Spatial,” a controversial award-winning Midjourney-generated image whose creator fought an extended battle to register it with the Copyright Office.

The office demonstrates the unpredictability of AI systems with a Gemini-produced image of a cat smoking a pipe and reading a newspaper, noting that Gemini ignored some prompt instructions and added a few things of its own — including an “incongruous human hand.” It contrasted this process with Jackson Pollock’s splatter painting method, where he didn’t control the exact placement of the paint on the canvas, but “controlled the choice of colors, number of layers, depth of texture, placement of each addition to the overall composition — and used his own body movements to execute each of these choices.” Ultimately, the office writes, “the issue is the degree of human control, rather than the predictability of the outcome.”

At the same time, the Copyright Office says that simply using AI to assist in human creative output does not necessarily jeopardize that work’s ability to be protected by the law. There’s a difference between AI being used as a tool to assist a creative work and “AI as a stand-in for human creativity,” and the office says that further analysis is warranted. But it assures creatives that using AI to outline a book or come up with song ideas shouldn’t impact the ability to copyright the final human-produced work, since the author is simply “referencing, but not incorporating, the output.”

Artists can get some protection if they feed their own work into an AI system for modification — by, say, using a tool to add 3D effects to an illustration. AI-generated elements of the work still wouldn’t be protectable, but if the original product remains recognizable, the “perceptible human expression” in the work could still be covered by copyright.

People can also receive protection for works that incorporate AI-generated content as long as there’s significant creative modification. A comic with AI-generated images can be covered if a human arranges those images and pairs them with human-written text, though the individual AI-generated images wouldn’t be protected. Likewise, “a film that includes AI-generated special effects or background artwork is copyrightable, even if the AI effects and artwork separately are not.” On a “case-by-case determination,” even prompt-generated images could be protected if a human selects and remixes specific areas of the picture. The office compares these scenarios to making copyrightable derivative works of human-created art — minus the original human.

A separate question is whether text prompts themselves can be protected by copyright. Overall, the office compared prompts to “instructions” that convey uncopyrightable ideas, but it acknowledged some particularly creative ones could include “expressive elements.” This doesn’t, however, translate into the work they produce being protected.

The Copyright Office didn’t rule out the potential for this to change if the technology evolves. “In theory, AI systems could someday allow users to exert so much control over how their expression is reflected in an output that the system’s contribution would become rote or mechanical,” the report says. But as of now, it doesn’t seem that prompts “adequately determine the expressive elements produced, or control how the system translates them into an output.”

This document is part of a larger effort by the Copyright Office to clarify policy questions and identify legal gaps around AI, starting with a July 2024 report encouraging new deepfake laws. The office next plans to issue a third and final report on its findings on “the legal implications of training AI models on copyrighted works.”