US rules AI-generated art can't be copyrighted

If you spent hours engineering the perfect prompt to get that photorealistic AI artwork, Uncle Sam has bad news for you.

The US Copyright Office (USCO) just published formal guidance in the Federal Register, confirming that images generated entirely by AI models like Midjourney or Stable Diffusion in response to text prompts cannot be copyrighted.

The commissioned artist analogy

The USCO's reasoning is surprisingly simple: typing prompts into an AI model is legally equivalent to a patron commissioning a human artist.

The user might describe the scene, mood, and style, but the machine decides how those instructions actually materialize. According to the office, the traditional elements of authorship are conceived and executed by the AI, not the person typing behind the keyboard.

Under US law, copyright requires human creativity. Because users do not exercise ultimate creative control over how these systems interpret prompts, pure text-to-image outputs are treated as mechanical reproductions.

The office even pointed back to its famous precedent involving monkey selfies, reinforcing that non-human creators simply do not qualify for copyright protections.

A crack in the door

The office made sure not to shut the door completely on works featuring AI elements.

Protection will depend on the specific circumstances of how the tool was used. If a human arranges, edits, or modifies AI-generated material in a sufficiently creative way, those human contributions can still be registered.

That nuance played out clearly in last month's decision on Kris Kashtanova's graphic novel, Zarya of the Dawn. The USCO ruled that the individual Midjourney-generated illustrations were not copyrightable, but Kashtanova retained copyright over the written story and the visual layout.

Kashtanova's legal team pushed back, arguing the office applied the wrong legal standard by scrutinizing the output rather than the creative inputs. To address the mounting debate, the USCO is hosting public panel discussions in April and May, with plans to gather public comments later this year.

Why it matters

This decision puts an immediate ceiling on the value of pure prompt engineering.

If your business model relies on churning out mid-journey assets and licensing them exclusively, your legal moat just evaporated. Anyone can legally take those raw outputs and use them commercially without paying you a dime.

To build protectable IP today, creators have to treat AI as a component rather than the finished product. Extensive retouching, manual compositing, and narrative structuring are now mandatory if you want legal ownership.

For AI toolmakers, the ruling creates a strong incentive to move beyond simple chat boxes. Companies will need to ship features that prove ultimate creative control—like precise inpainting, brush-level edits, and node-based workflows—so enterprise customers can defend their generated assets in court.