Copyright in the Era of Generative AI: Who Owns the Content?

Global Court Precedents and the Doctrine of “Human Authorship”

The central issue of modern jurisprudence has become whether an algorithm can be considered the author of a work. In 2023, the U.S. Copyright Office (USCO) issued guidance clearly stating that works created by a machine without substantial human creative input are not eligible for registration. A striking example was the case of “Zarya of the Dawn” by Kris Kashtanova. The regulator recognized copyright in the text and structure of the comic created by a human, but annulled rights to the images generated by Midjourney. This decision set a fundamental precedent: “prompt engineering” (writing text prompts) is not yet equated with traditional artistic creation, since the user does not control the final generated result with sufficient precision. In the court case of Stephen Thaler, who attempted to register a painting on behalf of the Creativity Machine algorithm, the U.S. District Court for the District of Columbia likewise ruled that human authorship is a “foundational requirement” of the law.

However, the situation is not so clear-cut when it comes to hybrid works. Lawyers emphasize that the key factor is the degree of human involvement. If an artist uses AI merely as a tool (for example, like a brush in Photoshop) and then makes significant manual changes, the work may receive protection. The problem lies in the absence of clear metrics: exactly what percentage of an image must be altered by a human? At present, courts assess each case individually, relying on the doctrine of de minimis-the minimum necessary level of creativity. This creates a risk zone for the corporate sector using neural networks in marketing and design, since such assets may effectively fall into the public domain and can be legally copied by competitors.

The economic consequences of legal uncertainty could amount to billions of dollars in lost revenue for creative industries. Studying the dynamics of patent disputes over the past year, independent financial analysts conclude that the lack of guarantees for intellectual property protection will slow long-term investment in generative AI by 15-20%. Companies may simply be afraid to invest resources in creating content that anyone could freely use tomorrow, which is why the market expects the emergence of new licensing models to insure against these risks.

Geography of Regulation: Polar National Approaches to AI Art

The world has split into two camps regarding the regulation of neural-network-generated content. In the United States and most EU countries, a conservative approach prevails, denying AI authorship. However, China demonstrates an opposite vector aimed at supporting the industry. At the end of 2023, the Beijing Internet Court issued a landmark ruling in the case of Li v. Liu, recognizing copyright in an image created using Stable Diffusion. The court argued that the operator invested intellectual effort in selecting parameters, choosing a model, and configuring prompts, which constitutes a form of “intellectual investment.” This decision makes China one of the most attractive jurisdictions for AI-content developers, creating a legal shield for digital artists.

The United Kingdom also occupies a special position thanks to the Copyright, Designs and Patents Act (CDPA) of 1988. Section 9(3) of this act explicitly addresses “computer-generated works” and recognizes as the author the person who undertook the necessary arrangements for the creation of the work. This is a unique legal construct that existed long before neural networks but has become especially relevant today. Unlike the United States, where a “human hand” is required, in the UK it is sufficient to prove the organization of the creation process, giving local design studios a significant advantage on the international stage.

Status of Copyright in AI-Generated Content by Country:

• USA: Not permitted (if there is no significant human post-processing). Content is considered public domain.
• China: Permitted (Beijing court precedent). Intellectual contribution in configuring the algorithm is recognized.
• United Kingdom: Permitted. “Computer-generated works” are protected; the system operator is recognized as the author.
• European Union: Not permitted (under the AI Act and current directives). Proof of personal human creative contribution is required.
• Ukraine: Partially permitted (the 2023 Law “On Copyright” introduces “sui generis rights” for non-original, computer-generated objects).

Boundaries of Liability and a Comparison of Human and Operator Rights

Delineating rights between a traditional author and a neural-network operator is not mere legal casuistry-it is a question of monetization. Even if a neural-network operator spends hours refining prompts, in most jurisdictions they do not obtain personal non-property (moral) rights, such as the right to attribution or the right to integrity of the work. Where such rights are recognized at all (as in the UK or Ukraine), they are usually limited to economic aspects-the right to authorize or prohibit copying. This places the operator in a more vulnerable position: the term of protection is often shorter (for example, 25 years instead of “life of the author plus 70 years”), and moral rights are virtually absent.

For businesses, this means a need to revise employment contracts. If a designer uses Midjourney to create a logo, the company must understand that the logo may not receive trademark protection. Experts recommend preserving all source files, generation logs, and-most importantly-layers of manual refinement in graphic editors. Only the presence of a “human touch”-such as redrawing details, compositing, or color correction-can transform “machine output” into a protectable object of intellectual property.

Comparative Table: Rights of the Author vs. Neural Network Operator