Artificial Intelligence and Intellectual Property Rights: Who Is Really the Creator?
Using artificial intelligence (AI) systems to make content like text, images, music, software code, and even scientific research raises one of the most important legal questions of our time: when does copyright apply to AI-generated output, and who, if anyone, is the creator? As generative AI tools become a part of almost every creative field, the gap between what technology can do and what the law says about human authorship has never been bigger.
In the field of AI, you can't just point to the tool or algorithm and say, "Who made it?" An AI model, no matter how advanced, is not a legal person. It has no rights, no responsibilities, and no intent. So, the answer must be found in the people who work with the AI and in the legal systems that need to change to fit a new reality that their creators never thought of.
The European Legal Framework: Originality as a Human Action
Copyright protection in European Union law necessitates an original intellectual creation stemming from human ingenuity. This principle is deeply rooted in the continental European tradition of droit d'auteur, which links the work to the author's personality—a concept that is fundamentally incompatible with machine authorship.
In a number of important decisions, the Court of Justice of the European Union (CJEU) has made this standard clear. The Court determined in Infopaq International A/S v. Danske Dagblades Forening (C-5/08, 2009) that copyright protection extends to works that constitute the author's "own intellectual creation." This standard was subsequently refined in Painer v. Standard VerlagsGmbH (C-145/10, 2011), where the Court asserted that a portrait photograph may be considered original if the photographer exercises free and creative discretion concerning framing, angle, lighting, and atmosphere. Under EU law, originality is closely tied to how the creator's personality comes through in the free and creative choices they make during the creative process.
In this context, when a Large Language Model (LLM) or image-generation model produces a result entirely and independently—without significant human involvement in the form, structure, selection, or final creative imprint—the establishment of copyright in favor of the user or the technology provider is neither apparent nor legally assured. In many places, this kind of output may automatically become public domain, meaning that anyone can use, copy, or change it without permission.
The Global Landscape: A Mix of Approaches
The problem is not only in the EU. All over the world, governments are dealing with the same basic problem, but their solutions are very different:
United States: The U.S. Copyright Office has made its position clearer and clearer. The Office gave copyright to the human-written text and the overall layout of the graphic novel Zarya of the Dawn in February 2023. However, it did not protect the individual AI-generated images made with Midjourney because the author did not have enough creative control over each image made. The Office has since published formal guidance (Federal Register, Vol. 88, No. 51, March 2023) confirming that works must be created by a human being to receive copyright while acknowledging that works containing AI-generated material may be protectable if there is sufficient human authorship in the selection, arrangement, or modification of the AI output. The U.S. District Court for the District of Columbia upheld the Copyright Office's decision not to register a work made by an AI system called DABUS on its own. This confirmed that U.S. copyright law requires that works be written by people.
UK: Is in a somewhat unique place. According to Section 9(3) of the Copyright, Designs and Patents Act 1988 (CDPA), if a computer-generated work doesn't have a human author, the person who made the arrangements for the work to be made is considered its author. This provision, which was first written with simpler automated processes in mind, theoretically allows copyright to exist in works made entirely by AI. This makes the UK one of the few places where this is possible by law. Nonetheless, the practical implementation of this provision in contemporary generative AI is predominantly untested in judicial contexts and is currently under continuous policy discussion.
China: Chinese courts have dealt with this issue in a number of important cases. According to the Beijing Internet Court in the Feilin v. Baidu case (2019), a financial report that an AI system made on its own was not eligible for copyright protection because it did not have the required human originality. But in a more recent ruling from 2023, a Chinese court said that an AI-generated image was protected by copyright because the user showed a lot of creative involvement in making the prompt and improving the output. This suggests that the level of human involvement is what matters.
Japan: Japan has been more lenient with training data. Article 30-4 of the Copyright Act in Japan says that machine learning can use copyrighted material in many cases. However, the copyright status of outputs made by AI is still mostly unclear.
The Human as Creator: What Makes a Good Contribution?
The important practical and legal difference is the kind and level of human creativity involved. Not all interactions with an AI system are the same; some are passive and some are very involved:
Situations that bolster the argument for human authorship:
Conceptual origination: Before using the AI tool, the person comes up with the main idea, theme, story structure, or artistic vision. The AI tool is then used as a sophisticated tool for execution.
Iterative refinement and curation: The human makes many outputs, chooses between them based on aesthetic or substantive judgment, combines elements, throws out versions that don't work, and guides later generations toward a specific creative vision.
Substantive editing and transformation: The person who edits, rearranges, annotates, or transforms the AI-generated raw material in a big way leaves their own creative mark on the final work.
Compositional arrangement: In works with more than one AI-generated part (like a graphic novel, a movie, or a multimedia installation), the choices a person makes about selection, sequencing, juxtaposition, and overall arrangement may be creative expression that is protected by law.
Control over parameters in both technical and artistic ways: In some cases, changing model settings, style parameters, seed values, inpainting, outpainting, and other generation controls in a very specific way may be the kind of creative decision-making that courts recognize.
Situations that undermine or invalidate the argument for human authorship:
One generic prompt with no other action: In most places, just giving a short, general prompt (like "write a poem about the sea" or "make a landscape painting") without doing anything else creative with the result is not likely to be enough for copyright. In this case, the output is mostly based on the model's training data and random processes, not on the user's creative personality.
Automated bulk generation: Using AI to make a lot of content without having a person look over, choose, or curate each piece of output.
Cosmetic or trivial changes: Small changes to the formatting, spell-checking, or surface-level changes to AI-generated output that don't change the creative content in any way.
Risk of infringing on the rights of others
Even though the question of who wrote the work is still open, there is another important issue: the rights of third parties whose works may have been used to train the AI model or whose works the AI output may accidentally reproduce or look very similar to.
Risks on the input side and training data: Most modern generative AI models are trained on huge datasets that include copyrighted material like books, articles, photos, music, and code. These datasets are often taken from the internet without the permission of the people who own the rights. This has led to a lot of high-profile lawsuits. The New York Times v. Microsoft and OpenAI (filed December 2023) says that OpenAI's models were trained on millions of copyrighted articles without permission and that they can copy large parts of those articles word for word. Visual artists (Andersen v. Stability AI, filed January 2023) and authors, including well-known novelists, have also filed similar lawsuits. The results of these cases will have a big effect on the law. The AI Act (Regulation (EU) 2024/1689, which went into effect in August 2024) and the Copyright Directive (Directive (EU) 2019/790, especially Articles 3 and 4 about exceptions for text and data mining) set up a partial framework in the EU. However, there are still many questions about what training uses are allowed and how well opt-out mechanisms can be enforced.
Risks on the output side: If AI-generated content copies or looks a lot like a protected work, the user, the deployer, or the provider of the AI system could be sued for copyright infringement, unfair competition, or violations of moral rights. Just because AI made the content doesn't mean that anyone is off the hook for it. Depending on the law in that area and the legal theory that applies, the user who published the output, the company that used the AI tool, or the developer who trained the model on infringing material may be responsible. The contractual terms of service, indemnification provisions, and changing regulatory mandates are all helping to manage the distribution of this risk over time.
What Contracts and Terms of Service Do
In the current state of legal uncertainty, contracts are very important for figuring out the rights and duties of everyone involved. Most AI service providers have long terms of service that cover:
Ownership of output: Some providers, like OpenAI under its current terms, give users rights to the output they create as long as they follow the rules for how to use it. Some people keep some rights or only give limited licenses. However, these contractual assignments cannot create copyright where none exists under applicable law; they merely allocate existing rights.
Liability and indemnification: Some providers will protect you against intellectual property claims that come up when you use their tools, while others will not.
Terms of service often limit how AI-generated content can be used. For example, they may not allow authors to misrepresent their work, use it in certain industries, or create certain types of content.
People and businesses that use AI-generated content for business purposes should carefully read these contract terms and think about whether they need more contracts, warranties, or insurance.
New rules and laws
The rules and regulations are changing quickly. The AI Act sets up a risk-based framework for AI systems in the EU and requires providers of general-purpose AI models to be open about their work. This includes having to publish summaries of training data and follow copyright law. Internationally, groups like WIPO (the World Intellectual Property Organization) have held talks about AI and intellectual property. Several national legislatures, including those in Brazil, India, and South Korea, are also looking into new laws that would apply only to works made by AI.
Final Thoughts
When it comes to AI, you can't answer the question of "who is the creator" by looking at the tool or algorithm itself. The main factors that affect it are the level and type of human creative input, as well as the rules and contracts that govern how the technology can be used. As AI technology gets better, the law will have to deal with more and more questions about who owns what, how to protect the interests of creators, users, developers, and the public, and whether new types of protection are needed to deal with the realities of machine-assisted and machine-generated creativity. Until then, anyone who makes or uses AI-generated content should know exactly what the risks and benefits are.