Abstract
This article examines the reasoning for and feasibility of creating sui generis protection for Artificial Intelligence generated works. It starts by exploring the challenges posed by AI to already-existing forms of intellectual property, and how it exceeds the boundaries of what they were created for. It looks at previous forms of sui generis protection, and distinguishes them from AI. It suggests potential provisions for AI-exclusive sui generis protection.
Introduction
The challenges artificial intelligence poses to intellectual property are unprecedented. Most developments in technology that revolutionized the process of creation merely transformed or streamlined it. Artificial intelligence can replace it entirely outside of a singular prompt.
The two most ubiquitous forms of IP protection are patents and copyrights. Patents grant the inventor the sole right to make, sell, use or import their novel, useful and non-obvious invention for 20 years. Patenting any form of software is disputed, since opponents perceive it as an abstract piece of literature (lines of code) rather than a tangible machine, comparable to a mathematical equation. Different jurisdictions have different approaches, but a largely used solution is to grant software patents strictly when they have a “technical effect”, or simply when the software can make some kind of alteration to hardware.
What it is and What is the issue?
Artificial intelligence, like other software, is built on algorithms and mathematical models, neither of which can be patented. The difference is traditional software is strictly an execution of code. AI derives training data but ultimately creates its own decision-making rules. Any input it gives can be considered an “abstract idea”. This means that any novel inventions AI generates without human input cannot be patented. But what if a sort of Large Language Model is created to find novel methods to design new inventions and increase the efficiency of those that exist? Can such models count as having a technical effect? Stephen Thaler created two inventions using his DABUS bot, claiming it barely needed his input, but neither could be patented since the court held only natural persons could hold patents.
The issues AI creates for copyright receive more attention: who can claim copyright for art or writing generated entirely by AI? Does it count as a tool or an author? Considering that AI is trained on existing work, can there be legal consequences for it being too similar to existing work, or does it qualify merely as inspiration? Considering that AI abstracts the data it is trained on into mathematical tokens, it cannot be quantitatively inferred whether how much it has truly “stolen” from the original author’s work. Additionally, copyright protects the expression of ideas and not the concepts or facts behind it themselves. It is hence difficult to fight a case for plagiarism when AI-generated work takes heavy inspiration from a copyright-owning author.
To summarize, AI exceeds the limits of existing protection. When intellectual property is defined as protecting creations of the human mind, what happens when a non-human mind “creates”? Traditional copyright regime frameworks are not equipped to deal with similar cases. The present approach states any work created entirely by AI cannot be registered as intellectual property. This can create further complications, such as people using AI to generate artwork and inventions, but claiming sole authorship. The two major positions are to either keep the current framework, where only natural persons can hold copyright, not including AI, or AI being recognised as an inventor and author. The issues with the first have been discussed, but giving AI traditional copyright protection has its own set of cons. AI’s output pulls from millions of different photographs, artworks, books, codes, theorems and existing inventions. Though it is debatable whether it has already reached such a stage, with its depth of knowledge and ability to learn, it has the potential to eventually challenge manmade work. Publishers, labels and others involved in the commercial aspects of creative work will likely mass produce AI-generated work rather than investing time and resources into creators with uncertain returns on investment. And though AI can produce a more comprehensive final product, the effort, imagination and synthesis of emotion required that for some, are what make any art valuable and affective, should never be replaced. Giving AI the right to patent ownership creates serious ownership complications and reduces liability in agreements.
Solution
A middle ground is creating sui generis protection for AI-generated work. Latin for “of its own kind”, sui generis protection is afforded to certain works for which traditional IP Protection is insufficient. Semiconductors, plant varieties and non-creative databases are some examples of works given sui generis protection in different jurisdictions. The difference is that all these examples do not fit into IP laws because of the product’s nature. Databases, or the arrangement of large amounts of data in a specific order, do not involve “creative expression” and cannot be copyrighted, and are not technical matter but the arrangement of abstract ideas, so cannot be patented. Semiconductors are the closest comparison, because some parts can be patented and copyrighted, but because of the complexity of their elements and rise of chip piracy, they necessitated a different regime for convenience. AI changes the process, but the output is the same. Is it possible to create a separate right for the origination of the work in itself and not the final result?
Suggested provisions create workarounds to specifically acknowledge the process while protecting the output. Shorter terms of protection (5-10 years compared to 20 years for patents) should be given, considering the speed at which AI can create. There should be a mark to distinguish work that is AI-generated above a certain predetermined threshold from that which is not, increasing consumer awareness and somewhat creating a separate playing field for AI -generated work. The qualifying attributes for registration such as originality and novelty could be mirrored, but worded considering context. For example, originality can be explained as needing to be transformative rather than a 1:1 reproduction of training material.
Conclusion
Artificial intelligence alters the longstanding relationship between intellectual property and human ingenuity. Effective legislation doesn’t try to erase that fact by blatantly excluding AI generated work from protection, nor does it make no distinction between the two. Legislation must find a middle ground between both.
References
Iglesias Portela , M. Shamuilila & Anderberg, A, ‘Intellectual Property and Artificial Intelligence – A literature review’ (2021) Joint Research Centre https://publications.jrc.ec.europa.eu/repository/handle/JRC119102
Benjamin Hardman & James Housel, ‘A Sui Generis Approach to the Protection of AI-Generated Works: Balancing Innovation and Authorship’ (2023) Social Science Research Network
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4557004
AS Bayındır & J Danaher, ‘Why We Should Recognise AI as an Inventor’ (2025) Journal of Bioethical Inquiry
https://pmc.ncbi.nlm.nih.gov/articles/PMC12575523/
M Aboy, K Liddell, A Lath, ‘Inventorship in the age of AI: examining the USPTO Guidance on AI-assisted inventions’(2025) Journal of Intellectual Property Law and Practice
https://academic.oup.com/jiplp/article/20/7/495/8089945?utm
IP India, ‘Sui Generis Nature of Protection’ (26 June 2026) https://ipindia.gov.in/sui-generis-nature-protection
Clarin, ‘Related rights and Databases’ (26 June 2026) https://www.clarin.eu/content/related-rights-and-databases
Pakyntein, K., & Mishra, D. R. K, The Semiconductor Layout Design Act, 2000 ‘Filling the IPR Gap in India’s Technology Regime’ (2025). International Journal of Law Management and Humanities https://ijlmh.com/wp-content/uploads/The-Semiconductor-Layout-Design-Act-2000.pdf?pdf=1