Vitaly Golovkov
This is one of the “hottest” and most tangled topics in modern law.
The problem is that the legal machinery simply cannot keep up with the speed of neural network development. If we break everything down, we run into three fundamental deadlocks:
1. The prompt as an “instruction” rather than an “object of art.”
From the perspective of classical law, a prompt is often viewed not as a creative work, but as a technical specification (design brief). If you tell an artist, “Draw me a ginger cat in a hat,” you do not become the author of the painting. The author is considered to be the one who embodied the idea. In the case of AI, the “performer” is an algorithm, which is not a legal subject. The result is a legal vacuum: the machine cannot hold rights, and the human (the prompt creator) has exerted insufficient “creative effort” to be considered a full-fledged creator.
2. Lack of uniqueness — the problem of prompt “theft” and proof.
A prompt can be copied (or extracted from metadata via “reverse engineering”), a button can be pressed, and a similar result obtained. Since AI produces different variations for the same prompt, proving that the second image is a plagiarism of the first is almost impossible. In court, visual images are usually compared. But if the prompt is the same and the images are different (even if in the same style), there is no direct theft of the “object.” There is theft of the “method,” and methods are not protected by copyright. Copying someone else’s request is a legal gray area.
3. Anthropocentricity of Law.
According to the laws of most countries (including the Russian Federation and the USA), an author can only be a human who has invested their creative labor. If a court decides that the human role was limited only to “selecting words,” the work may be recognized as unprotected “public domain” immediately at the moment of creation. This creates a paradox: the prompt author considers themselves a creator, but any person can take their result, use it for their own purposes, and the “author” will not be able to sue because they have no legally protected copyrights.
Lawsuits are already underway in the USA and Europe. Here are the main arguments and cases that reveal the problematic nature of legislative regulation of copyrights for created prompts and AI generation results:
The “Théâtre D’opéra Spatial” Case (Jason Allen). This is the most famous example. Jason Allen won an art competition with a painting created in Midjourney, but the U.S. Copyright Office refused him registration. The Office ruled that a human must make a substantial creative contribution. Simply entering text (a prompt) is not enough. The problem is that the author edited the prompt 600 times, adjusted weights, light, and composition, but this is not yet “creativity,” but rather “instruction for the machine.”
The Case of Kris Kashtanova and the comic “Zarya of the Dawn” (2023). Kris created a comic where the images were generated in Midjourney. Initially, she was granted copyright for everything, but part of it was later revoked. The U.S. Copyright Office ruled: the text and the layout of the frames (design) are protected because they were made by a human. But the images themselves are not, because they are “not the product of human authorship.”
The Case of Stephen Thaler and “A Recent Entrance to Paradise” (2023). Thaler attempted to register a painting by listing the neural network itself (Creativity Machine) as the author and himself as the owner of the rights. The District Court for the District of Columbia confirmed: copyright can only belong to a human. AI cannot be an author; therefore, it cannot transfer rights to a human. This closes the loophole of “I own the program, therefore I own everything it does.” No human creator — no protection.
The Argument of “Randomness.” The law protects what the author fully controls. AI has an inherent element of randomness (Seed). You enter a prompt, but you do not know down to the pixel what the result will be. Courts tend to believe that AI art is closer to a photograph taken by a “randomly” dropped camera than to painting. If you do not control the final brushstroke, you are not the author, but the “customer.”
The Problem of “Style” and “Method.” The legal fact is that copyright does not protect style. If I paint “like Van Gogh,” I am not breaking the law. A prompt is a style formula. If I copied your request “retro-futurism, neon, 8k, cyberpunk style,” I stole your “idea” or “method,” but the law only protects the specific “embodiment” (the final file). Copying a prompt today is an ethical theft, but not a legal one.
Will we arrive at “Prompt Law” or will AI content remain “no-man’s land” forever?
The term “Human Authorship Requirement” is the foundation upon which all current court refusals rest. Until the law changes and recognizes a “non-human” as an author (or the prompt as an object of law), any AI work hangs legally in the air. Legal thought is currently moving toward a division into:
Free-to-use AI Art: Unprotected content. The law still qualifies the prompt as a technical specification. If you ask a taxi driver to take you to an address via a specific route with several stops, that is not creativity.
Registered AI-Assisted Art: Works where the author records the entire creation process (prompt logs, sketches, edits) and thus receives a limited right to the prompt and/or the result.
What could become the trigger for the protection of Prompt Law?
1. The “Iterative Process” Doctrine. It is quite obvious that a simple prompt is not enough for authorship. But if you show the history of changes, it changes the matter. It is necessary to officially recognize the validity of the thesis: authorship can be proven through the number of iterations. This allows for discussion about the possibility of legislative regulation of the “Prompt Law” institution. For example, if you entered 500 or more clarifying prompts, changed weights, used Negative Prompts, and configured parameters (such as CFG Scale or Seed Control), you carried out intellectual activity during which you made decisions and creative choices. Exactly the same choice is made by the author of any work in the classical understanding of their creative contribution. A photographer who sets the light, focus, and shutter speed does not “draw” the photo; they manage the process, yet they are the author of the resulting shot.
2. The “Hybrid Method (Human + AI)” Doctrine. The most reliable way to prove the need to protect prompt creation work is manual “post-processing.” Thesis: pure generation is public domain — this is fair. But generation + subsequent retouching, collaging, or inpainting is already a derivative work. You take a “blank” from the AI and overlay your own unique layer (Overpainting). Here, copyright protects specifically your edits. The fact is that manual post-processing, recognized as an object of copyright, is no different from prompt post-processing.
3. The “Prompt as Unique Complex Code” Doctrine. It is difficult to dispute the argument that if a prompt is not just three words “draw a cat in a hat,” but a complex structure of 10,000 characters using unique combinations and original metaphors, it can in itself be considered a literary work.The problem is that even if we protect the prompt text, we do not protect the result of its operation (the image), as the link between word and pixel in AI is not direct and predictable. But this exactly suggests that protecting the prompt is possible and necessary. While we deal with the legal nature of the AI generation result, prompt protection is the minimum guarantee of copyright protection for prompt engineers.
Copyright protects a specific file, but not the method of its creation. The law protects the canvas and paints, but it is not ready to protect “intent.” A prompt is a method or pure intent. And until law learns to evaluate the intellectual weight of this intent, authors of AI content will remain legal ghosts We face a choice: either turn all AI art into “no-man’s land” (Public Domain) or recognize the prompt as a new form of intellectual code requiring protection. Authorship in the age of AI is not the right to own pixels. It is the right to the idea itself. And this battle for “intellectual territory” is only beginning.
Based on current global practice and patent office refusals, the author proposes forming a register of prompt copyrightability criteria, which can serve as a basis for introducing amendments to the legislation governing the protection of intellectual property.
As examples for discussion, let us consider the following:
Criterion of Creative Choice: A prompt must not be obvious or purely descriptive. The use of original metaphors, unusual stylistic combinations, and unique epithets that are not typical for the training of a given neural network is copyrightable.
Criterion of Engineering Complexity: The prompt must demonstrate a deep understanding of AI mechanics. The presence of Negative Prompts in the structure, which cut off standard neural network solutions, imposing the author’s will upon it.
Criterion of Iterative Development: This is the most important point for courts. Authorship is proven through the process, not through the result. The presence of prompt chain change logs, where it is visible how the author gradually made edits.
Criterion of Structural Originality: The prompt must be complex, using specific request construction algorithms (e.g., multi-layered description: light -> composition -> material -> emotional context -> technical camera parameters).
Criterion of Volume: Long, detailed requests (1,000 characters and more) are more often recognized as objects of copyright than short phrases. It is advisable to establish a minimum character threshold for a copyrightable prompt.
Criterion of Human Control: The ability of the prompt to provide a repeatable or predictable artistic result, minimizing the algorithm’s randomness factor. This is proof that the AI was merely a “performer,” not a “generator of ideas.” For example, the use of tools like ControlNet, where the author defines the skeleton of a pose or a depth map. This moves the prompt from the category of “do something” to the category of “the logical outcome of long work.”
The combination of these specific criteria forms a legally significant sign for recognizing a prompt as an object of exclusive rights (the result of intellectual activity) — the threshold of originality.
In order to implement proposals for determining the legal status of a prompt, the formulation of new yet understandable legal definitions is required. In the draft law concept proposed by the author, the following basic concepts are provided:
Intellectual algorithmic request (Prompt) — a unique sequence of text, numerical, or visual commands, expressed in the form of code or natural language, created by a human to manage generative artificial intelligence algorithms with the aim of obtaining a specific creative result.
Author of an Intellectual algorithmic request — a natural person through whose creative labor a prompt was created that meets the criteria for copyrightability.
Threshold of originality — a set of qualitative and quantitative characteristics of an intellectual algorithmic request (prompt) confirming that the result was achieved through systematic intellectual labor.
As a result of the above approach, the possibility opens up for recognizing a work created by AI as an object of copyright if it is the result of executing a copyrightable prompt.
Current legislation of the Russian Federation, international law, and the legal systems of the European Union, the USA, China, and other countries contain a critical legal vacuum: AI works remain in a “gray zone” due to the absence of direct human authorship in the final act of generation.
Existing copyright law is not fully adapted for artificial intelligence systems, as authorship in generative machine intelligence is proven through the process, not through the result, as is characteristic of traditional objects of intellectual rights.
We are slowly moving from passive observation of AI to active management of property rights in the digital environment, protecting the human creative factor in the era of the expansion of machine intelligence.
Dmitri Drobnitsky
Alexander Palladin
Robert Reich
