Portrait of the llama as a young artist.
It’s been a while since we covered the AI authorship debate here at Llama Towers, despite it being one of the first AI legal questions we discussed here nearly 10 years ago (time flies when you’re having fun, eh?). The gap has largely been because everyone’s attention has been centred squarely on copyright infringement, leaving the AI authorship question to take a backseat. While there have been some interesting isolated developments, nothing has warranted a full blog post, so I’ve mostly been sharing updates on social media.
It’s been nearly 3 years since generative AI burst into our lives, making this the perfect time to pause and assess the state of the field, particularly as recent developments may signal where things are headed. But before delving into the legal status of AI-generated works, it’s worth examining the broader landscape of generative AI, as several practical issues have begun to shape the entire discussion.
The state of generative AI
Perhaps one of the most interesting aspects of generative AI over the past few years has been the relative absence of copyright discussion around authorship. I believe this stems from several unexpected features of the technology’s rise.
The first observation is the remarkable advancement of these models, both in their ability to produce viable outputs, and in their widespread adoption. While there’s debate about actual usage levels – critics claim “nobody” uses them while enthusiasts suggest universal adoption – the reality is somewhere in between: lots of people are using genAI, while the tools haven’t been adopted by everyone. Nonetheless, AI tools are now embedded in countless devices, platforms and products, from Apple to Microsoft, making free generative tools widely accessible. This has inevitably led to the emergence of “AI slop”: millions of artificial social media interactions, hallucinated text in legal filings, and a flood of AI images of questionable quality across social media platforms.
Speaking from personal experience, I’ve integrated generative AI tools into significant aspects of my work, primarily creating images for class presentations and for this blog. I’ve also found tools like NotebookLM valuable for identifying relevant academic papers. Friends and acquaintances tell me they use generative AI for coding, email composition, sharing images with friends, creating memes and various other purposes. While precise usage statistics remain unclear, we can reasonably conclude that many people are at least experimenting with these technologies.
The adoption by the public has not been met by a similar level of adoption by the creative classes, at least not openly. Usually, any creator that admits to using AI openly can expect some social media backlash. However, as companies start to explore AI use more openly, I don’t think that this will be followed by more claims of authorship. What is happening is that for many people AI is just another technical tool at the disposal of creatives, but that its deployment is not just about writing a prompt and getting a ready-made product. What we are seeing is more of the incorporation of these tools into different stages of the creative process, and as such there is considerable amount of human intervention.
But if AI usage is as widespread as some of us suspect, where are all the lawsuits attempting to claim copyright over such works? We have only seen a few of them so far.
This brings me to the final point of note that we can see in AI development, and it is that so far those who are using these tools the most do not seem interested in claiming copyright over their creations. Using my own experience again as an example, I’ve now generated thousands of images using generative AI tools, and I have not felt any inclination to attempt to claim copyright over any of them, if you see one of the images I’ve used here and elsewhere, feel free to copy them at your leisure. There are quite a few people who call themselves AI artists, and while I do not want to enter the debate over whether AI art is actually art (more about it here), it seems to me that most of the people who have gotten very good at it aren’t particularly interested in claiming copyright over their creations.
So we are starting to see an interesting combination of factors that make it very unlikely that copyright authorship will become a more important legal issue in the future. The heaviest users aren’t interested in claiming copyright, while the commercial users are either not using the technology, or using it without disclosure. After all, there’s no copyright police checking the provenance of every piece of text or art being shared, except for some unfortunate cases of anti-AI sleuths engaged in witch-hunts that tend to accuse innocent people more often than not.
We’ll now look at developments in three jurisdictions.
Copyright authorship in some countries
China
The country with more case law in this area is China, where there have been 4 decisions exploring whether works generated by a AI can be subject to copyright protection; 2 cases predate the generative AI boom, and 2 are more recent.
In 2018, the Beijing Internet Court ruled on the case Feilin v Baidu, where the court considered a report partially generated by software. While the court emphasized human authorship as crucial for copyright protection, it recognized a degree of protectable interest in the human effort and investment involved, even if the work wasn’t fully copyrightable. The court ruled against the defendant for copying and altering the report without attribution, suggesting a possible move towards investment-based rights similar to database protection in the EU. The case highlighted the ambiguity of applying copyright to works with AI assistance, particularly when the software is primarily a data analytics tool.
A subsequent case in 2019, Tencent v Yinxun, further explored this issue. The Shenzhen court granted copyright protection to an article written by Tencent’s Dreamwriter AI, which produces numerous articles daily. Despite the defendant’s argument that AI-generated works lack human authorship and reside in the public domain, the court found the article original due to its structure, logic, and the “selection, analysis and judgement” of information. The court emphasized that the AI’s output reflected the plaintiff’s team’s choices and arrangements, effectively attributing the work’s creation to human direction. This decision marks a significant step, recognizing copyright protection for AI-generated works, particularly those produced by more sophisticated AI programs. (more about these two cases here).
In 2023, the case Li v Liu, a Chinese court ruled that an AI-generated image created using Stable Diffusion could be protected by copyright. The plaintiff, Mr. Li, had used the AI to generate an image and posted it online. The defendant, Ms. Liu, used the image in a blog post without permission, leading to a lawsuit. The court determined that the image met the criteria for copyright protection under Chinese law, emphasising Mr. Li’s intellectual contributions and aesthetic choices. The court also established that Mr. Li, not the AI service or its developers, was the author due to his significant intellectual input.
And finally, at the end of 2024 we have the case Chen v. Changshu Qin Hong Real Estate Development Co. A Chinese digital artist produced the image of “a red heart reflected in water, forming a complete heart shape” using Midjourney and Photoshop, and registered it with the Copyright Administration. The image was turned into a 3D animation by the defendants. The artist sued for copyright infringement and won. The court found that the image had copyright, but limited the protection to the 2D registered image as the 3D installation was not a direct reproduction. Defendants were ordered to pay 10k RMB ($1,364 USD) and issue apology.
So the Chinese courts now have found repeatedly that works generated by AI can be subject to copyright protection. The common denominator in the latest cases has been that a human author has done enough to prove they have an intellectual achievement, and therefore prove originality and therefore to be granted copyright protection. The need for sufficient human involvement is something that it’s going to be an important feature of how we move forward.
UK
The UK was the first country to address the issue, almost by chance. Section 9(3) of the Copyright, Designs and Patents Act (CDPA) states:
“In the case of a literary, dramatic, musical or artistic work which is computer-generated, the author shall be taken to be the person by whom the arrangements necessary for the creation of the work are undertaken.”
Furthermore, s 178 defines a computer-generated work as one that “is generated by computer in circumstances such that there is no human author of the work.” This is an elegant wording that makes settles who is the author of a work created by a machine, but it does not deal with the subsistence of copyright itself. However, this section has not been tested in court, in a feature that is now becoming quite familiar. The lack of applicability of a provision that has been in existence since 1988 should perhaps give us some clue as to what is going on, I think in large part by the points addressed at the top of this blog post. Similarly, the issue here is that s9(3) is about allocating authorship, the work still needs to meet the subsistence requirements, namely originality, and the test for originality in the UK is that a work has to an intellectual creation, and this would appear to necessitate some sort of human intervention. How much human intervention?
We got a glimpse of this in the case of THJ Systems Ltd v Sheridan [2023] EWCA Civ 1354. Lord Justice Arnold discussed whether a software developer owned copyright in a graphic user interface. He explained that the originality test in the UK is that a work has copyright if it is the “author’s own intellectual creation”, and that while this test sets the bar higher to the old “skill and labour” test, “even a simple portrait photograph” can satisfy it as long as it meets the requirements (at 23). Discussing what is the level of creativity needed for protection in a GUI and some charts, Arnold LJ comments that it “is plain that the degree of visual creativity which went into the R & P Charts was low. But that does not mean that there was no creativity at all. The consequence of the low degree of creativity is that the scope of protection conferred by copyright in the R & P Charts is correspondingly narrow, so that only a close copy would infringe.”
I’ve been an outspoken fan of s9(3) because I think that it solves the authorship problem, while also leaving the door open for having a case-by-case analysis of whether originality has been met. However, s9(3) may have its days numbered due to the latest consultation on copyright and AI, where the government has made it clear that it favours removing the section altogether.
Alas, if that is to be, we mus bid farewell to a wonderfully quirky and anachronistic legal provision, we hardly knew ye.
USA
The USA has been quite active in this area as it is a country where authors must register their work with the Copyright Office in order to enforce their rights in court. This has led to a large number of high-profile cases where the USCO has denied registration to several works generated with AI, these include “Zarya of the Dawn” by Kris Kashtanova, Dr Stephen Thaler’s “A Recent Entrance to Paradise“, and “Théâtre D’opéra Spatial“by Jason Allen. Thaler and Allen appealed the decisions, and these are at the moment of writing in the US courts, but I have to say that I do not think that the cases will be successful, perhaps Allen has a better shot, but it will depend on the facts of the case. Thaler already had a negative decision, which is under appeal.
These setbacks led everyone to assume that no AI works could be subject to copyright protection, after all, US law applies everywhere, right? (insert sarcasm tag here). However, not many people had noticed that the USCO had actually been registering “hundreds of works” generated with AI, according to Registrar Perlmutter. This happened where it was considered that human intervention in the creation of the work was enough to warrant copyright protection.
It is in this context that the Copyright Office published the second part of its report on Copyright and AI, this one dealing with copyrightability. The report followed a consultation and over 10,000 replies, and it seems to have upset almost everyone, but I think that it follows the practice that had already been taking place at the Office. The report concluded that current legal doctrines sufficiently address questions of copyrightability for works involving AI. A central principle is that human creativity remains essential for copyright protection: purely AI-generated content or content produced with minimal human control is not eligible for copyright. However, AI can be used as an assistive tool without affecting the copyright status of a resulting work, as long as a human author contributes protectable expression. In these cases, the human’s creative inputs—such as selection, arrangement, or modification of AI-generated material—can qualify for copyright protection.
The Office also clarified that prompts alone are not enough to constitute authorship, since they are considered merely instructions conveying unprotectable ideas. But if a user’s contributions go beyond prompts, the resulting work may be copyrightable if it reflects sufficient human creativity. Ultimately, the Office does not find it necessary to introduce new forms of copyright or legal protection for AI-generated content at this time, emphasizing that copyright exists primarily to incentivize human authors. It will continue to follow developments in technology and law, advising the public as needed.
We know that this is already being applied, so it will be interesting to see if this somehow is translated to the courts. The first test may be the ongoing Allen case.
Concluding
It’s been an interesting few years regarding AI copyright authorship, and I think that we are starting to see some sort of consensus starting to emerge in both policy and some case law. While everyone pretty much assumed early on that AI-generated works would not have copyright, we are starting to see that a more complex picture is emerging. Purely AI-generated works aren’t copyrightable for the most part, as there is no intellectual creation, human creativity, intellectual achievement, skill and labour, whatever originality standard you may want to use. But something that I am very glad to see is that if there is enough human intervention, enough of a creative spark, then the work could be protected. This includes complex prompts as has been shown in the Liu case in China, or demonstrating extensive editing and selection of outputs. Or as the USCO concludes, actions such as selection, modification, editing, and arrangement of outputs may be enough to grant copyright.
In the end what we could see is that generative AI will be seen as just another tool, one more akin to Photoshop and Word. Some works won’t have copyright, while others will. But that may not matter for most people. We are also entering a world of creative abundance, and that may change the way in which we think about copyright in the first place.
One thing remains clear though, I don’t think that any of my llama pictures may qualify for protection. Let’s see, what do I get when I type “A llama thinking about the future”? There’s only one way to find out.
