With the rapid development of Artificial Intelligence (AI) and its integration into every aspect of modern living, a wide range of questions and uncertainties have begun surfacing with respect to intellectual property (IP) protection. As AI system use is becoming widespread, not only as research tools, but as interactive collaborators helping inventors brainstorm, test concepts, or even generate technical solutions, their role in the creation and protection of inventions is increasingly opaque.
Ambiguities surrounding inventions developed with the use of AI tools are increasingly evident at every layer of the patenting process, including inventorship, novelty and disclosure, inventive step, jurisdictional law, philosophy, and ethics. In this opinion letter, we try to assess the impact of widespread AI use on patentability or, more simply put, if AI will make it harder or easier for mere ‘mortals’ to obtain a patent.
To simplify a complex system, the ‘holy grail’ of patentability revolves around the assessment of three basic criteria: (1) is it new, (2) is it non-obvious, (3) is it capable of industrial applicability (i.e. is it useful). A somewhat overlooked criterion, not related to the invention per se, is the requirement of a human inventor. This requirement is consistent across all major jurisdictions and was explicitly confirmed by the Legal Board of Appeal of the European Patent Office in the DABUS cases (J 8/20 and J 9/20). In these decisions, the Board held that the inventor named in a patent application must be a natural person with legal capacity, thereby excluding machines or AI systems from being recognized as inventors. Additionally, inventorship is tied to the act of conception in most legal systems. This conception relates to the formation of the inventive idea and not merely the execution of technical tasks. Therefore, a further challenge, even if AI systems could legally be named as an inventor, lies in determining whether AI-generated contributions cross the line from assistance into true conceptual insight.
From both the perspective of inventors and patent offices, the advantages of AI integration are undeniable. AI promises to increase efficiency, enhance research tools, automate patent and infringement monitoring, and enable the development of new business models. These advantages appear universal across all technological fields. As such, AI is becoming less of an optional tool and more of a foundational element in innovation.
Despite these developments, the criteria for assessing patentability will, at least for the foreseeable future, likely remain the same. To better understand how these dynamics play out in practice, we turn our focus to a specific field where AI is already making a profound impact: life sciences.
AI systems, including LLMs, are actively being used in many aspects of research and development, particularly in drug discovery. Drug discovery is the field of identifying new molecules or new uses for known molecules, such as therapeutic proteins like antibodies. The function of a protein is determined by its three-dimensional (3D) structure, which is, in turn, dictated by its unique amino acid sequence. This sequence drives the process of protein folding, where the chain of amino acids folds into a specific 3D shape essential for the protein’s biological role. Determining protein structures is a highly specialized field, traditionally conducted by crystallographers using methods that are both technically demanding and time-consuming.
To this end, a breakthrough occurred with the development of AlphaFold, a specialized deep learning system for protein structure prediction, for which its inventors were awarded the Nobel Prize in Chemistry in 2024. Its release represents a significant leap in our understanding of protein folding, and by extension, drug discovery. To put this into perspective, over 200 million predicted protein sequences and folded structures have now been released via the AlphaFold Structure Database. Separately, a further 2 million new crystal structures have been released in materials science databases. These numbers are expected to grow, along with the accuracy of their predictive capabilities, with each new iteration of these models.
Turning to the novelty requirement in the assessment of patentability, the sheer volume of AI-generated data introduces significant challenges in determining whether an invention is truly novel. For instance, once protein structures are disclosed in public databases, they enter the public domain and the structure alone can no longer be considered novel for patent purposes. However, a new application of such a protein, such as a new drug mechanism, delivery method, or diagnostic use, is still potentially patentable. Further, a previous AI-generated disclosure, such as a predicted protein structure, need not be enabling to destroy the novelty of a later patent filing. In contrast, a patent application must meet stricter enabling disclosure standards. This asymmetry between the standards applied to a novelty-destroying prior art document and to the assessment of an invention adds another layer of complexity to strategic planning in AI-driven R&D.
Conversely, it can be argued that AI enhances the discovery process by increasing the likelihood of identifying entirely new drug molecules or uncovering novel uses for existing compounds, thereby supporting the novelty requirement. However, this also raises concerns about potential monopolization of innovation by entities with access to the most advanced AI systems, large-scale datasets, and superior computing power, potentially concentrating patentable discoveries in the hands of a few dominant players.
Another emerging concern is the use of defensive publishing, where companies release large libraries of AI-generated compounds, often with minimal experimental validation, to pre-emptively block others from securing patent rights. If left unregulated, this strategy introduces greater uncertainty for researchers: a compound that required substantial investment to develop may already be buried in an unfiltered database of millions of structures, making it difficult to verify novelty or freedom to operate. Furthermore, this unpredictability can erode the commercial value of pursuing genuine experimental research, as patent protection becomes harder to secure for outputs that are costly to generate but easily invalidated by low-effort, AI-driven disclosures.
Conversely, companies may adopt the opposite extreme strategy: restricting the release of AI-generated data and instead seeking alternative forms of protection for their novel compounds, such as maintaining trade secrets or confidential know-how. These can be licensed or transferred similarly to patents, safeguarding commercial interests. However, their scope and boundaries are often less clearly defined, and enforcement and valuation can be more complex. In addition, these forms of IP have the additional societal cost of reducing the open sharing of knowledge, which could ultimately hinder innovation and scientific progress for the broader public good.
Turning to the assessment of obviousness, AI models introduce further complexity. Even if an invention, such as a drug molecule, is technically new, it may still be deemed obvious if a sufficiently similar or suggestive reference can be identified. The likelihood of such references being identified increases significantly with the scale and speed of AI-generated outputs. With large datasets of newly disclosed protein structures now available, a critical question emerges: how inventive is a drug that targets a protein whose structure is already known and predicted to work in a particular way for a particular disease?
The combined capabilities of human and machine intelligence may, in effect, raise the bar for the assessment of inventive step. Even under current standards, inventive step remains one of the most difficult criteria to satisfy in patent examination, typically judged from the perspective of a “person skilled in the art.” In an AI-augmented research environment, who exactly is this skilled person, and does this definition need to evolve? If this person is deemed to have access to increasingly sophisticated AI tools, would the threshold for an invention to be non-obvious require greater technical insight? Would AI tools be more likely to link prior art from human-interpreted distant fields, and thus render the combination of these more obvious to a skilled person deemed to have access to such tools?
While we cannot predict with certainty what the next decade, let alone a longer timeframe, will bring, it is clear that, as with any transformative technology, society must grapple with both the opportunities and the challenges posed by AI. Learning how to navigate these shifts effectively will be essential. It may well be that Pandora’s box has been opened, potentially triggering a surge in case law as courts are asked to interpret existing frameworks in light of novel, AI-driven inventions.
In the near term, we are likely to see continued growth in the number of patent filings related to AI-assisted discoveries. Over time, this may prompt a reassessment of the standards applied to patentability, particularly in areas such as inventorship, novelty, and obviousness, to ensure they remain fit for purpose in an AI-enhanced innovation landscape.
To this end, there are additional risks to anticipate, such as a surge in AI-generated patent applications potentially overwhelming examiners, degrading application or examination quality, and undermining the value of the patent system itself. In line with the rapid advancements in AI and LLMs, the patent system itself may evolve from what we currently understand it to be. Ironically, some of the concerns may be eased by the deployment of AI in supporting patent offices. Yet, sceptics argue that this may only worsen the underlying issues, leading to a black-box system of machine-to-machine communication. In this dystopian view, applicants may no longer be able to engage meaningfully with the decision-making process: arguments could become obsolete if the AI systems have already determined the outcome through logic inaccessible to human reasoning. Soon, we may simply have to accept the decision handed down to us by our new algorithmic overlords in a future rife with “computer-says-no” frustration.
In conclusion, the intersection of AI and intellectual property raises profound legal, technical, and philosophical challenges. While AI is opening new frontiers in discovery and innovation, it is also pushing the boundaries of existing legal frameworks. To ensure that human inventors remain meaningfully incentivized to innovate, there is an urgent need for clear regulatory guidance and well-defined frameworks that address AI’s evolving role in the patenting process.