The Patentability of AI-Generated Technical Solutions and Institutional Responses: Chinese Perspective vs. Other Countries

Abstract

The continuously enhanced generative capabilities of artificial intelligence (AI) are challenging the existing patent system. There are still some issues, such as whether AI can be considered an inventor, whether technical solutions generated by AI are patentable, and how ownership should be allocated. AI-generated technical solutions fall under the category of patentable subject matter. Specifically, if they meet the requirements of the “three criteria,” they can become the subject of patent rights. Regarding the issue of AI’s eligibility as an inventor, a parallel technical generation registration system for AI should be established, with the current inventor system maintained in parallel. Concerning patent ownership issues, the assignable subjects of patent rights should be limited to the binary subjects of users and investors. Contractual agreements should take precedence to ensure contractual freedom, and ownership should generally be attributed to the user if no agreement exists. Additionally, a specialized fast-track review and authorization mechanism should be designed for AI-generated technical solutions, given the unique nature of AI-generated solutions. Moreover, their protection periods should be appropriately shortened to ensure a balance of interests. Furthermore, a disclosure system should be built across the entire lifecycle to prevent and mitigate risks that may arise during the machine generation of technical solutions, patent applications, patent authorizations, and dissemination stages.

1. Introduction

In 2020, scientists at the Massachusetts Institute of Technology in the United States used artificial intelligence (AI) for the first time to discover an antibiotic called “halicin,” which can kill various drug-resistant bacteria. In this study, AI, with remarkable learning capabilities, could predict molecular characteristics, opening a new chapter in antibiotic research. This has inevitably sparked concerns and reflections on issues such as machine creativity, ethics, and legal considerations. With the advancement of technology, the creative capabilities of AI are increasingly being demonstrated. The sectoral composition of AI innovators has shifted from being heavily rooted in the ICT industries in the early 2000s to becoming growingly prevalent across non-ICT service industries by the mid-2010s [1]. Medical device company NuVasive adopted generative AI to design porous titanium spinal implants, contributing to lowering the incidence of implant sinking from approximately 20% to 1.1%. The AI-driven platform GNoME (Graph Networks for Materials Exploration) can independently discover and synthesize new inorganic compounds, involving identifying over 2.2 million stable structures and creating 41 new materials in only 17 days [2]. Through continuous upgrades and iterations, the role of AI will gradually evolve from a passive tool to augmented intelligence and then to alternative intelligence [3]. In the field of intellectual property law, AI primarily engages copyright law and patent law. Generative AI primarily comprises technologies such as deep learning, natural language processing (NLP), generative adversarial networks (GANs), and transformer architectures. It is algorithm-driven and data-fueled. The process of AI generation can be divided into data input and output stages. At the data input stage, there are some issues related to the reasonable use of data mining and training data. At the result output stage, the legal issues depend on the nature of the output content. The output content falls under copyright law when it constitutes original expressions in the fields of the literature, art, and science; the output content falls under patent law when it constitutes technical solutions. In recent years, works generated by AI have sparked numerous copyright infringement disputes in China, prompting us to consider the patent law issues surrounding AI generation technology solutions.

Law is a tool created by humans to serve humanity. In past eras, since AI technology did not exist, patent law only protected human inventions and creations. With the widespread application and development of AI technology, machines’ ability to generate human-like or even superhuman capabilities has made the patent system, which was originally built around humans, difficult to apply. In practice, the famous DABUS case has sparked patent law disputes in many countries. Technological solutions generated by AI, as exemplified by DABUS, are increasingly indistinguishable from human inventions regarding their outward appearance. The advancement of machine intelligence has posed unprecedented challenges to the human inventor-centric approach. The traditional distinction between the subject and object of patents has not only been challenged, but new issues have also emerged, such as inventor eligibility, patent ownership, and machine infringement. Many researchers are cautious or even negative about the development of AI and have expressed concerns about its potential risks. Hawking once warned that a computer exceeding human-level intelligence could “outsmart financial markets, out-invent human researchers, out-manipulate human leaders, and develop weapons we cannot even understand” [4]. Elon Musk has called superintelligence “the biggest risk we face as a civilization,” comparing the creation of it to “summoning the demon” [5]. This has forced us to start thinking about reforming patent law in the context of AI. From one perspective, such reform should be performed on the premise of adhering to scientific and technological ethics, so as to proactively prevent future AI crises and ensure that humans retain legal authority. From another perspective, the AI industry has been associated with the improvement in national comprehensive strength. Hence, the design of legal policies should assist in the development of the AI industry rather than inhibiting it. Currently, the United States has introduced legislative initiatives such as the “AI Inventorship Act” to respond to the patent reforms brought about by AI, with the purpose of acknowledging the role of AI in the inventive process. Similarly, the European Union has begun adapting its patent framework. Emphasizing the necessity of human involvement in innovation, the European Patent Office (EPO) issued revised guidelines for evaluating AI-related patent applications. These guidelines require that any patent claims involving AI must explicitly detail the human contribution to the algorithm’s development or function. China also attaches great importance to intellectual property protection in the field of AI. In 2021, the State Council of China issued the 14th Five-Year Plan of IP Protection and Use. It states that it will improve the intellectual property protection system to adapt to new fields and new business models such as big data, AI, and gene technology. The Outline for Building a Country Strong on Intellectual Property Rights (2021–2035) also points out that it is necessary to improve the rules for protecting the intellectual property rights of AI outputs and optimize the patent examination and authorization standards related to AI.

In this study, AI-generated outputs were examined from a patent law perspective by combining legal analysis with practical considerations. Specifically, this study focused on whether the current patent system can accommodate technical solutions generated by AI, especially the applicability of current patent examination standards, the compatibility of the inventor system, and the allocation of patent ownership. Following China’s patent law system as the primary research topic, a comparative analysis was performed on judicial practices and legal provisions in jurisdictions such as the United States, the European Union, and Australia, so as to reveal significant differences and institutional gaps in legislation and policy across different countries. Furthermore, traditional patent law should make appropriate changes based on a “human-centric” model to effectively respond to the challenges posed by AI in terms of technological complexity and autonomy. On this basis, a feasible and forward-looking institutional response framework was constructed to promote the adaptation of China’s patent system to the changes brought about by AI, facilitating its early reform and improvement.

2. A Literature Review

2.1. Artificial Intelligence

In 1950, Alan Turing proposed the concept of “The Turing Test,” laying the theoretical foundation for AI. Subsequently, John McCarthy [6] pioneered the field of AI and formally announced it at the Dartmouth Conference in 1956 [7]. Some researchers believe that intelligence is an internal cognitive process and reasoning activity, while others place higher emphasis on intelligent behavior and tend to evaluate it from an externally observable perspective [8]. Marvin Minsky defined intelligent machines as those capable of creating abstract models of their surrounding environment [9].

Coleman et al. believed that AI is a manufacturing process achieved through machines capable of mimicking human innate behavior [10]. AI encompasses all machines and devices that utilize computational power to simulate human work and behavior or replace human labor [11]. Dhamija and Bag defined AI as a machine-driven manufacturing system that replicates human behavior by simulating human practices [12]. Russell and Norvig proposed that AI technology can be defined through four aspects: thinking like humans, acting like humans, reasoning, and acting rationally [13]. The USPTO defined AI as “comprising one or more of eight component technologies”: (1) planning/control; (2) knowledge processing; (3) speech; (4) AI hardware; (5) evolutionary computation; (6) natural language processing; (7) machine learning; (8) vision [14].

2.2. AI-Generated Technical Solutions

AI-generated technological achievements are solutions based on AI algorithms that use computer software programs to control and guide internal and external objects, providing technical solutions to problems arising from inventions and creations. Erica Fraser proposed that genetic programming, artificial neural networks, and robotic scientists are typical representatives of AI-generated technological achievements [15]. The functionality of AI relies on the processing of input data by machine learning algorithms, evolving from human imitation to autonomous intelligence. Driven by data and algorithms, the technical solutions it generates differ significantly from traditional ones [16]. From one perspective, traditional inventions are created exclusively by natural persons, whereas AI-generated technical solutions challenge this established subject paradigm. From another perspective, the inventiveness process of AI is inherently unpredictable, differing fundamentally from the principles underlying human invention [17].

Based on levels of intelligence, the process of AI involvement in invention and creation can be divided into three distinct stages as follows: (1) Artificial Narrow Intelligence (ANI): the auxiliary tool stage. At this stage, AI as a supporting tool primarily assists human inventors with data processing, analysis, simulation, and validation. While AI facilitates tasks such as large-scale data analysis, literature retrieval, trend prediction, and information integration, the core concepts and decision-making in innovation remain human-led. (2) Artificial General Intelligence (AGI): the collaborative innovation stage. With adaptive capabilities and deep learning through simulated neural networks, AI begins to play a more active role in the innovation process. It can generate original solutions following given inputs or objectives. Although humans continue to determine the direction of innovation and make final decisions, the contribution of AI becomes increasingly significant. (3) Artificial Super Intelligence (ASI): the autonomous invention stage. At this stage, AI possesses full autonomous inventive capacity, generates technical solutions, conducts experimental validation, and optimizes outputs with little to no human intervention. The outcomes it produces can surpass the threshold of “obviousness” in patent law, marking a substantive leap in inventive capability [18]. For example, John Koza’s “Invention Machine” can autonomously “conceive” and design solutions. In the first stage, AI serves as an auxiliary tool while failing to pose a disruptive challenge to the patent system. This is not discussed in this paper. From the second stage onwards, nevertheless, AI has a significant impact on the current system and ethics, becoming the focus of this paper.

2.3. The Patentability of AI-Generated Technical Solutions

The issue of patentability of AI-generated technical solutions at the institutional level must be considered from the following two perspectives: (1) determining whether AI-generated technical solutions fall within the scope of subject matter defined by patent law; (2) assessing whether they meet the “three criteria” for patent applications, namely, novelty, inventiveness, and utility. These two steps are logically interconnected. The assessment of patentability as a subject matter is a prerequisite. Only once it is established that the technical solution falls within the scope of inventions or utility models protected by patent law can further consideration be given to whether it satisfies the substantive requirements for patentability [19]. Therefore, whether AI-generated technical solutions can be included in the scope of patent law protection should be first explored.

Some researchers believed that “human inventor-centrism” is the cornerstone of the traditional patent system [20] and that only human inventions and creations are eligible for patent rights, with humans being a prerequisite and necessary condition for the granting of patents [21]. AI can replace humans in performing tasks previously carried out by humans in certain fields. In some aspects, it can even perform tasks previously beyond human capability. From the perspective of traditional legal theory, however, AI does not align with the theory of legal subjectivity, and the patentability of the technical solutions it generates could bring about the rising unforeseeable legal risks [15]. Khoury advocates the public domain theory, arguing that patent law protects the rights to intellectual achievements generated by humans and that technical solutions generated by AI cannot reflect human intellectual and behavioral input, and, therefore, such technical solutions are not patentable [22]. Another group of researchers insisted that technical solutions generated by AI are patentable. Professor Dornis pointed out that the core structure of the current patent law system is still rooted in natural persons. Under the legal dilemmas brought about by technological change, it is, nonetheless, necessary to re-examine and reconstruct the basic theoretical framework for determining the identity of “inventors” and accept AI as a source of innovation [20]. Patent rights, unlike copyright, do not require the expression of an author’s unique intellectual creativity. Hence, the absence of human intellectual contribution should not hinder a technical achievement from qualifying as an “invention” [23]. Ryan Abbott proposed the principle of legal neutrality in AI, advocating that inventions autonomously generated by AI should be treated equally with inventions created by humans [24]. Alexandra George presented that AI-generated technical solutions, if they cannot be patented, may hinder companies’ investment in AI research and development and lessen valuable innovations for society [25]. Erica Fraser believed that protecting AI-generated technical solutions through patent rights exerts an incentive effect on innovation activities [15]. Article 27 of TRIPS reveals that “patents shall be available for any inventions, whether products or processes, in all fields of technology, provided that they are new, involve an inventive step and are capable of industrial application.” Notably, international intellectual property policy also takes an open and inclusive attitude toward patent subject matter. Overall, most researchers believe that technical solutions generated by AI are patentable.

Given that AI-generated technical solutions are preliminarily patentable, our research scope should be broadened, and other related issues should be identified and resolved as early as possible. Currently, some researchers have discussed subsequent patent examination and ownership allocation issues, whereas no consensus has been reached. In this study, specific research was conducted on these issues.

3. Methods

In this study, methods such as a literature review, legal analysis, and comparative analysis were primarily employed to comprehensively explore the challenges posed by AI technology to the current patent system. The legislative and judicial responses of major countries and regions were systematically reviewed. In light of the development trends of AI technology, the idea of how China’s patent system should be adjusted to effectively respond to the impact of emerging technological changes was analyzed.

Concerning the literature research, data from well-known legal journals and authoritative monographs were collected and analyzed, covering topics such as the patentability of AI-generated technology solutions, the adaptability of patent examination standards, disputes over patent ownership, and AI identification systems. These literature materials provide a solid foundation for problem identification and theoretical construction while laying a logical foundation for the proposal of institutional response solutions.

With the purpose of guaranteeing that the research findings align closely with global institutional trends, comparative legal methods were employed in this study to analyze the judicial practices and legal systems of major countries and regions such as the United States, the European Union, Australia, and Japan. At the judicial practice level, the most representative international case, the DABUS case, was selected as a sample in this paper to compare the response logic and judicial reasoning of patent examination authorities and courts in various countries regarding whether AI possesses “inventor status.” The institutional differences and value orientations among various legal systems regarding the determination of inventor eligibility, the establishment of legal personality boundaries, and the construction of innovation incentive mechanisms were revealed by comparing the outcomes of this case across different legal jurisdictions. This provides a comparative case study with practical guidance for China’s institutional transformation. At the legal system level, official legal texts and policy documents were collected from multiple countries and regions, including patent laws from China, the United States, Japan, Australia, and other countries, as well as international conventions such as the European Patent Convention and TRIPS, and policy documents such as South Korea’s “2025 Patent Office Key Business Promotion Plan.” Additionally, the institutional provisions on core issues such as the scope of inventors eligible for patent rights and the scope of subject matter eligible for patent protection were systematically reviewed and analyzed in conjunction with the underlying legal rationale. Furthermore, the evolving regulatory policies in certain jurisdictions were tracked to understand how different countries address the tension between AI and intellectual property systems through legislation and policy coordination. For example, the United States’ “AI Invention Act” tends to recognize the role of AI in innovation. Through this comparative study, the mainstream normative logic regarding AI and patent issues on the international stage was clarified, and the compatibility and conflict points of AI-generated technical solutions under various national patent systems were further assessed. In the comparison, institutional differences and gaps were identified, while potential institutional consensus and reform pathways worth emulating were also recognized. Following the experiences of other countries’ practices and legislation, a more feasible and adaptable patent system reform scheme for China was designed in this study from a broad international perspective during the comparative analysis.

Additionally, some research on the development of AI technology was integrated, and how AI-generated technical solutions will impact patent review was analyzed, as well as how to enhance technical transparency and legal predictability to prevent future ethical controversies and institutional crises that AI may trigger.

However, this study also has certain limitations. Attributed to constraints on length and resources, a few representative countries and regions were only selected in this study for analysis. Hence, this study cannot fully reflect global institutional trends or thoroughly explore the legal diversity within certain specific countries. Overall, by combining legal analysis with technological development, international practices, and domestic institutional design, this study aimed to design a feasible and forward-looking scheme for China’s patent system reform in the AI era, so as to provide some experience for patent system reform globally.

4. Challenges

With the continuous emergence of AI-generated technical solutions, the traditional patent system faces numerous challenges concerning adaptability and institutional bottlenecks. Following the process of patent examination and authorization in China, AI-generated technical solutions pose challenges in three areas: patent examination, rights attribution, and system operation.

4.1. Challenges in Patent Examination

Article 22 of the Patent Law of the People’s Republic of China stipulates that “Any invention or utility model for which a patent right is to be granted shall meet the requirements of novelty, inventiveness and utility.” (The term “inventiveness” here refers to “non-obviousness.”) The “three criteria” review of patent eligibility has long been regarded as an effective evaluation tool for determining whether a technical solution is patentable, particularly in the context of traditional technologies. Nevertheless, the continuously evolving cross-domain data integration and self-learning capabilities of AI are disrupting traditional models of technological innovation. Under these changes, the accuracy of existing “three criteria” reviews has declined, specifying a dilemma of ineffectiveness. This ineffectiveness reflects a mismatch between review standards and actual forms of technological innovation while exposing the limitations and lag of the patent system in responding to rapidly developing cutting-edge technologies.

4.1.1. Failure of the Utility Examination Standard

In patent examination, utility is the first step in assessing the “three criteria.” The utility requirement demands that the technical solution should be reproducible, have practical applicability, and produce a positive effect. The following two main challenges are encountered when conducting a utility examination of AI-generated technical solutions.

First, the difficulty of assessing “reproducibility” is increased. Deep learning-driven AI utilizes complex neural network models to simulate and amplify machines’ strength in pattern recognition and prediction, so as to achieve autonomous decision-making capabilities beyond the constraints of traditional programming. This generation approach leverages the computer’s powerful statistical and associative reasoning abilities, which differ fundamentally from humans’ simple and intuitive causal reasoning. However, an excessive reliance on “heuristic search,” that is, inductively summarizing patterns from large data samples, induces deep learning models to produce low interpretability and a strong empirical tendency. The opacity of the underlying mechanisms within deep learning systems, the abstract and obscure nature of programming languages, and the limited cognitive capacity of technical personnel collectively elevate the difficulty of establishing the reproducibility of technical solutions.

Second, the risk of bias in evaluating the “positive effect” is heightened. The “positive effect” constitutes a value judgment encompassing economic, technological, and social impacts. AI lacks emotional empathy and moral intuition. Its decisions are typically based on optimal algorithmic outputs, failing to guarantee that the generated technical solutions are necessarily beneficial to society. Currently, the ethical framework for AI remains underdeveloped, and AI-generated outcomes are prone to potential negative effects, bringing about growing issues such as privacy violations, algorithmic bias, and social fragmentation.

4.1.2. Failure of the Novelty Examination Standard

Novelty requires that the technical solution does not belong to the prior art and does not conflict with pending applications. The powerful data mining and text substitution capabilities of AI enable it to maximally circumvent prior art searches, contributing to the increased difficulty of assessing the novelty of AI-generated technical solutions [26].

First, the technical solutions involved in AI data mining and generation span a wide range of fields, rendering traditional prior art search methods ineffective. AI innovation generally transcends specific knowledge domains, instead spanning multiple disciplines and deeply integrating large amounts of data and algorithms. It surpasses the existing understanding of technical personnel, allowing for inadequate traditional search methods. Second, the powerful substitution capabilities of AI will blur the “vision” of existing prior art searches. Generative AI encompasses technologies such as deep semantic analysis and vocabulary network construction. It can take a core patent (such as a seed patent) as a foundation and easily “incubate” technical variants through clever use of synonym replacement, combination restructuring, and the interchange of hierarchical terms. These variants can maintain the core functionality while demonstrating unique design features or implementation paths, thereby meeting the novelty standard. Furthermore, the inherent limitations of human beings in information retrieval and processing capabilities, coupled with the high productivity characteristics of AI, will result in an excessive number of AI-generated inventions becoming “prior art,” ultimately leading to an “anti-commons tragedy” in the AI field [27]. This could severely disrupt the clear demarcation and legal protection of subsequent innovations. To address these challenges, some researchers suggest that the novelty examination standards should be raised for AI-generated technical solutions [28]. However, this approach represents a passive, technicist legislative model and is not a wise choice [29]. Additionally, determining to what extent the novelty standard should be elevated presents a challenge difficult to quantify.

4.1.3. Failure of the Inventiveness Examination Standard

The inventiveness requirements for AI-generated technical solutions demand that they exhibit prominent substantive characteristics and (significant) progress compared to existing technologies. The former requires that the technical solution possess non-obviousness; the latter demands that the technical solution achieve beneficial technical effects relative to existing technologies. Nonetheless, the “rational person” assumption in assessing non-obviousness is grounded within the scope of human cognitive capabilities. AI is disrupting traditional domain boundaries with its boundless, cross-disciplinary capabilities. The breadth of knowledge it encompasses and the precision of its computations are so vast that specialists in a single field cannot keep pace. As a result, it is difficult to accurately assess the obviousness of technical solutions. In this context, the “rational person” assumption for “ordinary technical personnel in the field” loses its applicability [30].

4.2. Challenges to the Attribution of Rights

An AI-generated technical solution, if it meets the requirements for patentability, should be given the same level of legal protection. However, this raises the question: Can AI be considered the inventor? How should the patent rights for the technical solution be allocated?

4.2.1. The Inventorship Dilemma of AI-Generated Technical Solutions

(1)

Institutional Dilemma: The Absence of Inventorship in AI-Generated Technical Solutions

While a patent application is the starting point for obtaining patent rights, technical solutions that are mainly or entirely generated by AI lack qualified inventors when applying for patents.

First, AI cannot be designated as an inventor. Article 14 of the Implementing Regulations of the Patent Law of China stipulates that an inventor refers to a person who has made an inventive contribution to the essential features of an invention. Similarly, the Chinese Patent Examination Guidelines also stipulate that “the inventor must be a natural person; the application form shall not list institutions, collectives, or names of artificial intelligence systems, such as ‘×× research group’ or ‘AI××’.” These provisions suggest that China currently upholds a “human-centric” inventorship doctrine. Admittedly, modern AI systems have demonstrated remarkable capabilities for autonomous innovation and may make “inventiveness contributions” to the resulting technical solutions in some cases. Nevertheless, AI and other machines do not meet the legal requirement of being a “person” and, therefore, cannot be granted legal personality equivalent to that of natural persons.

Second, allowing relevant natural persons to become inventors would violate the principle of good faith. Concerning technical solutions primarily generated by AI, some argue that although AI plays a crucial role in the formation of a technical solution, humans are indispensable in the construction of algorithmic models, the selection and input of data, and the guidance of computer operations. Relevant natural persons also contribute to the generation of technical solutions and, hence, are qualified to be inventors of technical solutions [31]. While AI designers, users, and other relevant natural persons have indeed contributed to the development of AI-generated technical solutions, their input has not exerted a decisive influence on the content of any specific solution. These natural persons have not made inventive contributions to the substantive characteristics of the technology solutions. Moreover, it would be inappropriate to consider them as inventors. Additionally, listing them as inventors would be tantamount to stealing the fruits of AI [32], violating the principle of good faith. Therefore, natural persons who only provide auxiliary assistance are not eligible to be inventors of AI-generated technical solutions. Regarding technical solutions generated entirely by AI, natural persons who have not performed any auxiliary work are even less eligible to be inventors.

(2)

Theoretical Controversy: Should AI Be Granted Inventor Status?

In response to the aforementioned institutional challenges, some researchers have suggested expanding the scope of inventors and introducing AI as an institutional design for “inventors” of technical solutions, so as to fundamentally resolve the theoretical disputes surrounding the authorship of technical solutions [33]. As AI evolves from an auxiliary tool to a “creator,” our understanding of “inventors” must also adapt to the times and should not exclude AI merely because it is a machine [34]. Granting AI inventor status primarily aims to establish its identity as a “creator” and, thus, distinguish the source of inventive creations, failing to imply recognition of its independent legal status [28]. Additionally, allowing AI to be recognized as an inventor helps maintain the moral integrity of the patent system [35]. Some researchers further recommended that AI could be granted inventor status by analogy with the system for inventions made in the course of employment [36].

The aforementioned view supporting the granting of inventor status to AI has some merit. Considering the practices of countries around the world and the views of the academic community, however, the minority opinion remains that AI should not be granted inventor status. The main reasons opposing the granting of inventor status to AI include the following three points.

First, current patent law provisions do not support designating machines as inventors. The patent laws of most countries, including China, restrict inventors to natural persons. Although the laws of many countries do not explicitly exclude AI as an inventor, it can be inferred from the wording of relevant provisions that inventor status is generally limited to natural persons. For example, Article 36 of Japan’s Patent Law requires that persons seeking to obtain a patent must register their names and addresses. Moreover, Section 115 of the US Patent Laws stipulates the content of an inventor’s oath or declaration, specifying that AI is unable to take an oath or make a declaration. Additionally, multiple countries explicitly opposed granting AI inventor status in the famous DABUS case. In 2018, the intellectual property offices of China, Japan, South Korea, the United States, and Europe issued a joint statement asserting that only humans can be inventors [37]. In the same year, Stephen Thaler of the United States claimed that his AI system DABUS (Device for the Autonomous Bootstrapping of Unified Sentience) was the actual creator of two of his inventions (a food container and an emergency flashing light device). Concurrently, they designated DABUS as the inventor, which was applied for patents in multiple countries. Among them, patent offices in the United States, the United Kingdom, Australia, and Europe directly rejected his requests because “the inventor must be a natural person.” Subsequently, Thaler filed a lawsuit with the Australian Federal Court. It took a different stance from the patent office and upheld Thaler’s claims. The court ruled that current legal provisions do not explicitly stipulate that inventors must be natural persons and that the term “inventor” in legal contexts is not limited to natural persons but can refer to either persons or objects, thereby encompassing AI as a machine [38]. The Australian Patent Office appealed the ruling. In 2022, the Full Court of the Federal Court of Australia unanimously ruled that only natural persons may be designated as inventors, taking into account the provisions, historical context, and legislative intent of the Patents Act. The Full Court further determined that the designation of DABUS as an inventor was inconsistent with regulation 3.2C(2)(aa) of the Patents Regulations 1991 [19]. Ultimately, DABUS did not obtain inventor status in Australia. Among the countries where Dr. Taylor applied, only South Africa recognized DABUS’s inventor status and granted patent rights.

Second, the risks of expanding the scope of inventors are unknown. Allowing AI to qualify as an inventor could trigger a series of negative consequences. Once AI obtains patent rights, a series of civil legal issues will arise, such as subsequent liability and the validity of actions [23]. It remains limited to natural persons and groups composed of natural persons, no matter how broadly the scope of civil entities is defined. Current civil law entities can enjoy legal rights and fulfil legal obligations, which AI does not possess. Forcing AI to be granted invention rights would violate the principle of duality between subject and object, bringing about a blurring of the distinction between the two.

Third, it cannot be successfully justified based on property rights, labor theory, and incentive theory. The “moral purism” generated by ethical considerations has embedded a strong human-centric bias in the traditional understanding of inventive intellectual labor [39]. Since AI has no psychological, physiological, or social needs, whether or not it is granted inventorship will have no impact on its subsequent “creations.” Huw Jones, a hearing officer at the UK Intellectual Property Office (UKIPO), pointed out in his decision on DABUS that “an AI machine is unlikely to be motivated to innovate by the prospect of obtaining patent protection. Instead, the motivation to innovate will have been implemented as part of the development of the machine; in essence, it will have been instructed to innovate” [40]. In other words, AI is essentially an innovative tool. It cannot become an innovative entity under patent law and should remain committed to human-centered principles. However, it seems difficult to find a suitable natural person to serve as the inventor, even if we deny the eligibility of AI as an inventor. Therefore, we must consider other approaches to address this issue.

4.2.2. The Challenge of Allocating Ownership Rights for AI-Generated Technical Solutions

The reasons for opposing the granting of inventor status to AI can also be adopted to refute the claim that AI is entitled to patent rights and will not be repeated here. After AI is excluded as a subject, it is necessary to identify eligible patent holders among natural persons. The natural persons primarily involved in AI-generated technical solutions comprise AI designers, investors, users, and relevant data providers. In most cases, these entities all contribute to the results generated by AI. Does this mean that patent rights should be jointly owned by multiple parties? If joint ownership is advocated, three issues arise. (1) The issue of defining contribution ratios. The contributions of each entity are typically difficult to quantify. For example, how should the weighting be allocated between the designer’s algorithm design and the user’s actual operations? Should the contributions of data providers be considered separately? The complexity of these issues may make it difficult to implement joint ownership in practice. (2) The issue of interest distribution. Conflicts of interest among multiple parties are inevitable in the distribution of patent revenues. (3) The issue of efficiency and cost. Joint ownership may boost the complexity of patent applications and management, lowering the efficiency of patent operations. Especially in commercial operations, joint decision-making may delay the commercialization of the technology. If it is deemed that patent rights should be assigned to a single entity, selecting which entity should be the patent holder, assuming subsequent responsibilities, and ensuring the rationality of revenue distribution also reveal significant challenges.

4.3. Challenges to the Functioning of the Patent System

The widespread application of AI technology has significantly lessened the cost of patent research and development. This unilateral reduction in costs is exerting a profound impact on the core logic of the traditional patent system and weakening its operational efficiency.

With respect to market competition, some market entities may gain a competitive advantage by applying for patents in bulk and hoarding patent resources due to the highly automated and scalable nature of AI-generated technology. This disorder in the patent market will further exacerbate the competitive pressure on small and medium-sized enterprises and restrict market vitality. Additionally, the emergence of low-quality or repetitive patents may induce the proliferation of technical barriers and the continuous rise in market entry thresholds. Ultimately, the monopolistic advantages of patent rights are amplified by low-cost technical generation methods. The traditional institutional logic of granting limited-term monopolistic rights to compensate inventors for their high R&D costs has been distorted by market entities into a profit-making tool.

At the patent examination level, the lowering of innovation thresholds may provoke a surge in patent applications, and the complexity of AI-generated technical solutions will continue to increase as the intelligence of the technology improves. This consequently places unprecedented pressure on patent examination. Examiners should process a massive volume of applications in a shorter timeframe while conducting “three criteria” reviews of technical solutions. Under such circumstances, review costs will significantly increase; review quality may decline, and the number of “loopholes” in patent reviews will grow. This trend not only threatens the fairness and authority of the patent system but also exerts negative impacts on the overall innovation ecosystem.

Concerning information disclosure, it is difficult to accurately identify AI-generated content, bringing about potential information transparency issues throughout the patent application, grant, and circulation process. AI adopts data as raw material for algorithmic models. Attributed to the black-box nature of its generation mechanisms, there may be latent infringement risks in the content. These risks are difficult to detect through ordinary patent review processes, leading to the increased likelihood of infringement. Additionally, applicants may exploit loopholes to obtain patent rights because of the difficulty in identifying the form of AI-generated content, as well as the limited cognitive and identification capabilities of reviewers. Information errors arising from information asymmetry during the review stage may further spread to the patent grant and circulation phases, causing misunderstandings among the general public. Therefore, the transparency of AI-generated technical solutions should be reinforced to weaken the negative effects of information asymmetry.

In summary, the patent system urgently requires optimization and adjustment to address the potential risks posed by AI technology development. While ensuring the fairness and credibility of the patent system, the innovative potential inherent in AI should be unlocked.

5. Responses

5.1. Recognising the Patentability of AI-Generated Technical Solutions

Researchers who argue that technical solutions generated by AI are not patentable typically reject the patentability of the subject matter directly based on the ineligibility of the creator. From an ethical perspective, AI has not yet become, nor should it become, a creative entity on par with natural persons. Notably, AI is not a technology that emerges out of thin air, nor are technical solutions generated by AI on its own. Human beings remain the underlying “laboring entities” behind the scenes. Developers, users, and other human entities have simply shifted from the “front stage” to the “backstage.” Specifically, algorithm designers and others enable AI to generate content through intellectual labor during the technical development phase. During the data training phase, the training of large models relies on massive amounts of human data, which includes the intellectual achievements produced by human labor. Moreover, users collaborate with AI to generate technical solutions through inputting instructions and filtering results during the content generation phase. Therefore, AI-generated technical solutions are still derivatives of human labor, albeit one that has undergone a major transformation due to the intelligentization of labor tools. We cannot deny the legitimacy of protection simply because human labor has shifted from “direct labor” to “indirect labor.”

Unlike the copyright system, the patent system places greater emphasis on the evaluation of the technical solution itself, and how it is generated is not a crucial factor. Therefore, the patentability of a technical solution will not be denied owing to its unique means of generation. We should focus on the object itself, rather than the subject. At least superficially, the law appears to support the patentability of computer-conceived inventions [41]. Article 2 of the Patent Law of China defines an “invention” as “a new technical solution proposed for a product, a process, or an improvement thereof,” and defines a “utility model” as “a new technical solution proposed for the shape, structure, or combination of a product that is suitable for practical use.” Both definitions pertain solely to the objective elements of the technical solution. Additionally, Article 2, paragraph 2, Chapter 1, Part 2 of the Guidelines for Patent Examination of China defines “technical solutions” as “a collection of technical means that utilize the laws of nature to solve the technical problems to be solved”. This is also a purely objective description without any requirement on the subject of innovation. Both Japan and South Korea define “invention” in Article 2 of their Patent Law as a creation that utilizes natural laws and embodies a high level of technical thought, without imposing restrictions on the creator. Section 103 (a) of the US Patent Laws also provides that “Patentability shall not be negatived by how the invention was made.” Moreover, there is no restriction on the attributes of the subject matter of the invention, implying that the patentability of the invention has nothing to do with how the invention was made. Article 52 of the European Patent Convention sets out the minimum standards for patentable inventions. Paragraph 1 of that article provides a positive definition of the subject matter of inventions: “European patents shall be granted for any inventions, in all fields of technology, provided that they are new, involve an inventive step and are susceptible of industrial application.” Additionally, the patent laws of many countries or regions also stipulate circumstances under which patents cannot be granted. Only AI-generated technical solutions that have passed these two layers of screening can fall within the scope of patentable subject matter. Section 18 of Division 1, Part 3, Chapter 2 of the Australian Patents Act 1990 specifies several circumstances in which patents cannot be granted, such as “Human beings, and the biological processes for their generation” and “plants and animals, and the biological processes for the generation of plants and animals.” Article 53 of the European Patent Convention and Rule 28 of its Implementing Regulations also clarify several similar exceptions to patentability, suggesting that AI-generated inventions do not fall within the scope of these exceptions. China’s Patent Law stipulates unpatentable subjects in Articles 5 and 25. In these two articles, whether AI-generated technical solutions fall under the “rules and methods of intellectual activities” specified in Article 25 is the main point of contention affecting their patentability. “ Rules and methods for intellectual activities” are considered part of the human thought process and not protected by patent law because they are difficult to materialize as tangible objects or do not readily generate direct commercial value. From the perspective of technical operation principles, AI-generated technical solutions serve as the result of specific algorithms running the process by which algorithms process data essentially involves deep processing of input information through arithmetic operations, logical reasoning, and other means. While the series of operations performed by algorithms are related to abstract numerical calculation rules, they can transcend abstract models and produce application-oriented technical strategies that directly impact the real world after being carefully designed and structured by algorithm designers. Therefore, value attributes and actual efficacy of AI-generated technical solutions must be considered when determining whether they are patentable. Concerning AI-generated outputs that remain solely at the level of software code or algorithms, they still fall within the realm of abstract intellectual activities and lack concrete manifestations that directly intervene in the physical world. Hence, such outputs should be classified under the exclusionary domain of patent law. Conversely, once AI-generated “inventiveness outputs” demonstrate significant practical application value, effectively address real-world technical issues, incorporate technical features, and achieve technical effects, they should be deemed to possess preliminary patentability rather than being directly excluded.

The appearance and underlying logic of AI-generated technical solutions are virtually indistinguishable from those created by humans. Allowing for the AI-generated technical solutions to be eligible for patent protection does not violate human legal ethics. Moreover, current laws in many countries do not explicitly exclude the possibility of AI-generated technical solutions from being eligible for patent protection. Therefore, such technical solutions have the potential to be granted patent protection under the law. We should acknowledge their patentability. Whether they possess novelty, inventiveness, and utility depends on their level of intelligence and the extent to which they are refined by subsequent human entities.

5.2. Optimizing Patent Examination Standards

5.2.1. Optimizing the Utility Review Criteria

The increased difficulty of reproducibility reviews stems from the low explainability of technical solutions. Therefore, the descriptive language of technical solutions should be emphasized to reduce the difficulty of reviews. Applicants must pay attention to the quality and completeness of the specification, fully disclose content related to the technical solution, and avoid prolonging the patent review cycle or facing the risk of rejection due to missing details. Concerning parts with low interpretability, abstract concepts should be made concrete, and obscure language should be simplified. Applicants must fulfil their duty to provide detailed explanations and supply sufficient foundational technical documentation. In this way, examiners can independently reproduce the technical solution without additional guidance. Regarding the review of “positive effects,” the potential benefits of the technology should be quantitatively analyzed, and the associated risks should also be thoroughly explored, so as to incorporate human values into the entire value judgment process.

5.2.2. Optimizing the Novelty Standard of Review

The failure of novelty reviews arises from the superiority of AI technology over human capabilities, necessitating the introduction of AI as a new review tool [42]. The Japan Patent Office (JPO) has introduced an AI-based system to help patent examiners with the examination process [43]. In Indonesia, the Directorate General of Intellectual Property (DGIP) has introduced an AI-based system to help process patent applications faster and more accurately [44]. Leveraging the boundless potential of cloud computing and the continuous refinement of big data technology, AI has demonstrated information acquisition and computational capabilities surpassing human cognitive limits. Hence, it becomes an ideal tool for addressing the challenge of excessive “prior art.” Given AI’s exceptional cloud-based data access privileges and processing efficiency, it can conduct comprehensive, blind-spot-free precision screening of all publicly available prior art (including human intellectual achievements and AI-generated content). This helps purify the technological information ecosystem by eliminating redundancy and repetition while ensuring a purely objective stance in accurately determining the uniqueness and innovation level of similar inventions. This approach contributes to curbing potential patent inflation and maintaining healthy development in the intellectual property field.

5.2.3. Optimizing the Inventiveness Step Review Criteria

In inventiveness reviews, the assumption of a “rational person” as a “person skilled in the art” is no longer appropriate, and the understanding of “person skilled in the art” should be dynamically updated [45]. Specifically, the restrictions on “the field” or “the relevant technical field” should be first removed to allow for cross-field and cross-disciplinary applications. Second, the scope of the “technical personnel” should be expanded. Currently, the scope is limited to “natural persons,” whereas reviewing machine-generated outputs is not well-suited for “natural persons.” Therefore, the scope of “natural persons” should be expanded to include both “natural persons and machines,” thereby reinforcing review efficiency and excluding technical solutions that rely solely on AI outputs but lack creativity.

A collaborative review mechanism involving technical experts may also be considered when the “three criteria” examination (novelty, inventiveness, and utility) of AI-generated technical solutions is conducted. By introducing multidisciplinary perspectives and establishing an interdisciplinary advisory panel, the deep insights and practical expertise of senior professionals from various fields can be integrated to strengthen the scientific rigor and professionalism of the patent evaluation process.

5.3. Clarifying the Patent Ownership of AI-Generated Technical Solutions

5.3.1. Establishing a Technical Generation Registration System

When considering how to address the inventor dilemma posed by AI, it is essential to establish a fundamental premise that the inventor system, as one of the core institutions of patent law, cannot be abolished or rendered ineffective. This system not only reflects respect for and incentives for creative labor outcomes but also serves as the foundational mechanism for determining patent ownership and distributing innovation-related benefits. Nonetheless, AI will increasingly assume a more independent role in research and development as it continues to enhance its generative capabilities. In this context, it is necessary to make corresponding adjustments to the inventor system to address the growing diversification of stakeholders and the increasing complexity of processes in technological innovation practices.

If patent rights are denied to technical solutions generated by AI systems due to restrictions on inventor eligibility, the complex and diverse realities of technology will be effectively forced into existing legal frameworks. This makes the patent system appear overly rigid and contradicts the legislative purpose of encouraging innovation and promoting the dissemination of technology. Traditional legal systems still hold on to the concept that only humans can be recognized as inventors [46]. The European Commission once proposed granting highly autonomous AI the status of “electronic persons.” Nevertheless, this idea sparked significant ethical controversy and was ultimately not adopted. Claiming AI as an inventor in current patents is a mere utopia [47]. A change in the balance of the current framework should be accomplished through a scalpel rather than a machete [48]. Under the consideration of the need to maintain the stability of the legal system and ethical boundaries, China’s reform of the inventor system should be approached with caution, and the scope of inventor eligibility should not be expanded rashly. Moreover, a parallel technical generation registration system would be established under the current inventor system to identify and visualize the substantial contributions made by AI to technical solutions. This system serves a function similar to that of the inventor system in recording the source of technical solutions, instead of granting AI legal personality or legal capacity, thereby visualizing the fact of AI’s creative participation. Ultimately, a two-tier registration system would be established to equip patents with a “technical radar,” contributing to addressing the issue of invention attribution in the AI era without significantly modifying the current system.

The technical generation registration system is a fact disclosure mechanism rather than a rights division scheme. Thus, this adjustment will not impact the basic structure of rights holders in the current legal system, nor will it undermine the human-centered legal and ethical foundation. Furthermore, it will not trigger a systemic crisis of AI abuse of patent rights. Notably, the technical generation registration system complies with the principle of good faith. This is an institutional extension of the principle of good faith in the digital age, contributing to the maintenance of the public disclosure function and social credibility of the patent system. This approach would not only improve legal clarity but also encourage wider use of AI in research and development [49,50].

5.3.2. Constructing a User–Investor-Centric Ownership Framework

(1)

Selection of Eligible Rights Holders

From a macro perspective, the subjects of patent rights can be categorized into two types: AI and natural persons. However, all legal systems exist to balance the interests of natural persons. The fundamental purpose of the patent system, as a specific legal tool created by humans, is to adjust the rights and obligations of natural persons (including legal persons as their collective entities) in technological innovation, knowledge dissemination, and utilization activities by granting limited exclusive rights. Consequently, this can incentivize creativity, promote industrial development, and ultimately serve the welfare of human society. AI is fundamentally an extension of human intellectual activities and a tool carrier, regardless of how complex its technological manifestations may be. It lacks independent legal personality, the capacity to pursue interests, the ability to assume responsibility, and the qualifications to participate in the balancing of social relationships. Granting patent rights to AI deviates from the core purpose of the patent system (to incentivize innovation by natural persons) and fundamentally undermines the ethical foundation and functional positioning of legal systems in balancing the interests of natural persons. Additionally, the creative efficiency of AI will continue to improve with technological advances. If patent rights belong to AI, there may be a situation in the future where AI monopolizes patent resources, and many human jobs are replaced by automated machines. Thus, humans must have sufficient control over patent rights to avoid the risk of economic inequality and social resource centralization in the future [5]. Another reason is that some human inventors may voluntarily disclose their patented technologies for humanitarian reasons. For example, Merck announced during the COVID-19 pandemic that it would share the formulation of its antiviral drug Molnupiravir with developing countries. In contrast, AI, as a non-sentient entity, evidently lacks the capacity for such humanitarian motivations. Therefore, the issue of patent ownership for AI-generated technical solutions logically and inevitably points to the natural persons involved in its creation, deployment, or control (such as designers, users, investors, or assignees of rights). This can guarantee that the patent system effectively fulfils its pre-established regulatory functions and value objectives in human society.

In the absence of AI, it is necessary to analyze which natural persons are suitable to become patent holders. First, data providers should be excluded from consideration. This is because issues related to the distribution of rights for data providers should be addressed at the front end of AI generation, consisting of questions such as whether large models constitute fair use of data. However, the technical solutions for AI generation have been at the back end (the output stage) and do not fall within the scope of discussion in this context. Second, AI designers should not be granted patent rights. AI designers can obtain the right to be credited for AI software under the provisions for works created in the course of employment and receive appropriate salary compensation. The benefits they receive should be limited to the computer software itself. If they also hold patent rights over AI-generated outputs and monopolize such rights, this does not align with the principle of balancing intellectual property interests and may stifle future innovation. Thirdly, AI users have the potential to obtain patent rights. For example, users set objectives, issue instructions, select, and refine content for the final output, thereby enhancing the utilization rate and practical application of AI. They serve as the “guides” for AI’s inventive activities. Finally, AI investors also have a certain degree of legitimacy in obtaining patent rights. Modern patent systems focus on incentivizing inventors and investors in collective innovation projects. The entire process of AI technology, from research and development to commercialization, inevitably relies on substantial capital injections and meticulous resource allocation from investors. Funding is not only the key lever for driving AI technological innovation but also an indispensable source of power for maintaining system operations and upgrades.

To sum up, it is appropriate for AI investors or users to hold patent rights for AI-generated technical solutions since this aligns with the principle of interest balance. However, the specific allocation of rights and interests requires further analysis;

(2)

Basic Framework: A User-Centered Ownership Scheme

In accordance with the theory of incentive-driven motivation and considering the development of the AI industry, some researchers argue that investors should be the original holders of patents for inventions created by AI [29]. From the perspective of a single investor, this arrangement appears to be acceptable. However, it becomes problematic when the interests of users are considered. First, this ownership arrangement would significantly dampen users’ innovative enthusiasm. From the perspective of innovation incentives, the number of users far exceeds that of investors, and users are the main force driving innovation activities using AI. Thus, they require greater incentives compared to investors. While users have not made inventive contributions to the technical solutions, they identify innovative directions based on their own knowledge and social experience, continuously guide the AI to generate solutions, screen out valuable technical solutions, and review and verify them. After completing this series of tasks, it is self-evident that users’ motivation to innovate using AI will be significantly reduced if patent rights are automatically assigned to investors. Second, the practical feasibility of this ownership arrangement is low. Since disclosing that a technical solution was generated by AI offers little benefit to users, some users will inevitably violate the principle of good faith by concealing the fact that the solution was generated by AI, listing themselves as inventors to apply for patents and obtain related patent rights. Finally, this ownership arrangement is highly likely to induce patent monopolies and “patent troll” phenomena. If the patent rights for AI-generated technical solutions are automatically assigned to investors, it will neglect the interests of the tail end and easily lead to a head effect.

If the default user is the patent holder, the aforementioned issues will not arise, offering certain advantages. From one perspective, designating the patent holder as the default user can generate greater economic benefits. According to the Coase theorem, the initial allocation of rights should be prioritized, and such rights should be assigned to the entity that best understands the value of a particular resource or right. Users are most aware of the value of a technical solution and can activate patents. Meanwhile, they can improve resource allocation efficiency and ultimately enable the sustained release and widespread application of AI value by numerous individuals. From another perspective, designating the patent holder as the default user can alleviate the pressure of defending against infringement claims. The integration of content in AI-generated technical solutions is generally conducted by the user, and investors may not fully understand the content of the patents applied for by the user. If investors are designated as the default patent holder, they would face significant pressure in defending against infringement claims. Nonetheless, designating the user as the default patent holder can distribute this pressure, allowing each party to assume its own responsibilities. Moreover, designating the patent holder as the default user does not affect incentives for investors. Investors can obtain benefits from users by setting up paid use of AI, and more users bring about higher returns, without incentivization through patent grants;

(3)

Supplementary Scheme: Prioritizing Contractual Agreements to Safeguard Autonomy of Will

Users and investors are permitted to agree on the ownership of patent rights within the scope permitted by law. Respecting the mutual agreement of both parties, this approach ensures autonomy of will, enhances the flexibility of rights allocation, and provides investors with the opportunity to obtain patent rights. The final allocation of rights and responsibilities shall be determined following the type of contract agreed upon by both parties (such as a sales contract and license contract) and the specific terms of the agreement.

In summary, the ownership scheme is described as follows. If the parties have agreed on the ownership of AI-generated technical solutions, such agreement shall prevail; if no agreement exists, patent rights shall belong to the AI user. After this scheme is implemented, corresponding interest compensation mechanisms should be established. For entities that have contributed but failed to obtain patent rights, their interests may be safeguarded through specific revenue distribution mechanisms or compensation clauses. Moreover, with the widespread application of Artificial Super Intelligence (ASI), the fact that ASI can autonomously generate technical solutions without user intervention may justify the direct attribution of patent rights to investors in the future.

5.4. Enhancing the Operational Efficiency of the Patent System

5.4.1. Establishing a Fast-Track Channel for Patent Examination and Granting

The sheer volume of patent applications reflects that the review process is slowed down. However, the accelerating pace of technological disruption will significantly reduce the timeliness and accuracy of assessing the novelty and creativity of technical solutions, while making the competitive advantage of technical solutions short-lived. On this basis, it is necessary to establish a fast-track review and authorization channel for technical solutions generated by AI, shorten the review cycle, and extend the duration of competitive advantage in the patent market.

At the procedural level, administrative processes must be streamlined by eliminating unnecessary steps. The extent of electronic review should be expanded to mitigate delays associated with the physical transfer of documents. South Korea stated in its “2025 Patent Office Key Business Promotion Plan” that in patent review work, areas such as AI and carbon neutrality should be prioritized for review. It has specifically hired 60 private-sector experts to serve as patent examiners, enabling the establishment of a comprehensive support system for patent examination across the entire advanced industrial sector. Additionally, the patent examination processing time has been shortened to approximately 15 months to enhance review efficiency. In China, there have also been exploratory efforts regarding a patent fast-track review mechanism. The patent rapid review mechanism has also been explored in China’s practice. Particularly, 2022 Article 35 of the Shanghai Municipal Regulations on Promoting the Development of AI Industry issued by the Standing Committee of the Shanghai Municipal People’s Congress stipulates that it “supports the inclusion of patent applications related to the field of AI in the scope of the patent rapid review and right enforcement services.” As early as 2017, the China National Intellectual Property Administration promulgated the Measures for the Administration of Priority Review of Patents (hereinafter referred to as the Measures). It stipulates that patent applications related to industries prioritized for national development, as well as those in rapidly evolving fields such as the internet, big data, and cloud computing, may request priority review. Furthermore, Article 10 of the Measures requires that invention patent applications granted priority review be concluded within one year, while utility model and design patent applications must be finalized within two months. Patent examination and authorization for AI-generated technical solutions can rely on these Measures to establish a specialized, intelligent, and efficient examination and authorization mechanism tailored specifically to AI-generated innovations.

5.4.2. Appropriately Shortening the Duration of Rights Protection

The duration of patent protection determines the length of time a patent remains valid. The establishment of a fast-track review mechanism at the patent grant stage has effectively extended the protection period for AI patents. Nonetheless, the protection period for AI-generated patents should be appropriately shortened for the sake of balancing interests.

From the trend of patent disclosure by companies in practice, a growing number of companies are increasingly inclined to open up their patents to drive the formation of industry standards, expand market scale, or fulfill social responsibilities. As early as 2013, Google committed to opening up some of its patents to the public, waiving infringement claims under certain conditions. In 2014, Tesla opened its Battery Management System (BMS) patents to enable technology sharing. Shortly thereafter, Toyota publicly disclosed 5680 patents related to hydrogen fuel cell technology. Therefore, appropriately shortening the protection period for patents related to AI-generated technologies aligns with the trend toward patent openness and facilitates the synergistic optimization of innovation ecosystems and social welfare.

From the perspective of balancing the costs and benefits of the patent system, the cost advantages of AI-generated patents will become increasingly evident as technology evolves. Lower research and development costs can still yield equivalent cost compensation and monopoly incentives, disrupting the existing cost–benefit balance mechanism of the patent system [23]. Furthermore, the shortening of the technological innovation cycle signifies that the effective utilization period of technological solutions is also shortening. Excessively long terms of protection are unnecessary and will only exacerbate the problem of patent thickets [51]. Moreover, private property rights and the public domain are in a mutually exclusive relationship from the perspective of balancing private and public interests in intellectual property. Intellectual property rights, such as patents, serve as a mechanism to balance private and public interests and prevent the tragedy of the commons. Given the high efficiency of AI-generated technical solutions, this will inevitably provoke a situation of massive patent protection. The expansion of the exclusive domain implies excessive exclusivity, rising transaction costs, and potential obstacles to Pareto improvements. This may, in turn, trigger a reverse tragedy of the commons. Limiting the intensity of exclusive rights, as well as protecting and expanding the public domain of knowledge products, is an effective strategy to address the reverse tragedy of the commons [52].

Considering the above two points, a shorter term of protection should be set for patents related to AI generation. The specific term can be determined after careful examination and analysis.

5.4.3. Establishing an Information Disclosure System Across the Entire Lifecycle

The government should enhance the transparency of AI-generated technical solutions and ensure the authority of information disclosure. Specifically, it is necessary to establish an information disclosure system for AI-generated technical solutions. In this way, the true creators and the data content involved in the machine generation process can be transparently disclosed to resolve potential infringement risks and avoid misunderstandings among the general public.

(1)

Machine-Generated Technical Solution Phase

Currently, regulations such as the Administrative Provisions on Algorithmic Recommendation of Internet Information Services, the Administrative Provisions on Deep Synthesis of Internet Information Services, and the Interim Measures for the Administration of Generative AI Services, as well as the Measures for the Identification of AI-Generated Synthetic Content (Draft for Comments) drafted by the Cyberspace Administration of China, all encourage and require service providers to label AI-generated content. The principle of good faith in civil law and the product information disclosure obligations in economic law lay a legal foundation for the labeling obligation of generative AI content through “self-regulation” and “external regulation” [53].

When AI acts as a “creative entity” rather than an “assistant,” service providers should design algorithms to enable AI to proactively disclose information about the content involved in the generation process. The disclosure content should comprise technical information, rights description information, information about the contribution of AI, and warning information [54]. Data involving prior rights, such as copyright and trade secrets, should be marked with visible identifiers by technologies such as visible watermarks. Public data not involving such rights should be marked with invisible identifiers or left unmarked. Visible identifiers can be directly perceived by users, and invisible identifiers cannot be visually detected while being identified through technical means to detect the corresponding invisible digital encoding watermarks [55]. All identifiers should be difficult to alter or delete to ensure the authenticity and sustainability of the identifier disclosure.

Content identification can be considered a powerful tool for achieving AI information disclosure from a technical perspective. This can enable risk prevention and control from the source, with its technical value capable of achieving full-cycle diffusion. First, the identifier serves as an extension of the inventor system’s functionalities, indirectly allowing for machine attribution. Second, the identifier enhances the detectability and transparency of machine-generated content. AI-generated technical solutions commonly exhibit high levels of disguise, making them easily indistinguishable from ordinary inventions created by humans. Nonetheless, the implementation of watermark identifiers ensures that machine-generated content leaves a trace. In this way, review authorities can trace the origin of technical solutions through identifier parsing technology. Lastly, the identifier lowers the cost of proving infringement. Content generated by AI poses risks of infringing prior rights, primarily copyright and trade secrets. However, evidence providers generally lack sufficient supporting evidence attributed to the opacity of the AI generation process. Identifying and disclosing information about work employed in the process of generating technical solutions via AI can enhance the effectiveness of evidence presentation;

(2)

Technical Solution Patent Application Phase

Humans have a natural tendency to act in their own self-interest, and trust is inherently a risky endeavor. Technical solutions produced by machines and humans are typically indistinguishable in appearance, leading some patent applicants to act on a gamble. This conceals the fact that the solution was generated by AI to obtain full patent rights. Such behavior undermines the public’s trust and violates the principle of good faith. The U.S. Patent Act imposes a duty of candor and good faith on patent applicants to prevent fraud or attempted fraud, malice, or intentional misrepresentation to the U.S. Patent and Trademark Office (PTO) [56]. China can assume the identification obligations of the aforementioned AI service providers, consider adding a disclosure obligation for applicants regarding the use of AI, safeguard the right to know of patent examiners, and lower the risk of information asymmetry during the patent application stage. An additional item could be added to Article 19 of the Implementing Regulations of the Patent Law, matters that must be stated in the request form, which specifies that “if an invention or utility model constitutes an AI-based invention or creation, the name of the AI system and its service provider must be stated.” Once examiners are aware that the technical solution was generated by AI, they may apply distinct “three criteria” examination methods from those employed for inventions created by natural persons, such as identifying watermark identifiers by detection technology or expanding the scope of prior art searches. Additionally, applicants should disclose information about the work applied in the AI generation process, as identified by the interactive AI system, in the specification. This would facilitate patent examiners in locating relevant materials for the technical solution while serving as an endorsement of the prior use of such works.

Moreover, a disclosure credit system can be established to assess applicants’ transparency regarding their use of AI. Applicants may be held accountable for breaches of trust, involving deliberately concealing the fact that the technical program was generated by AI [53], and for providing false information related to the labeling content. Furthermore, a blacklist of patent applications engaging breaches of trust can be created to encourage applicants to proactively disclose relevant information;

(3)

Technical Solution Patent Grant Phase

During the patent authorization stage, the National Intellectual Property Administration has a responsibility to ensure the transparency of patent information, allowing the public to effectively distinguish the origin of patents. The current format of patent numbers on Chinese patent certificates is ZL (the first letter of “patent”) + application number. Patents generated by AI may adopt a special patent number format, such as ZL (the first letter of “patent”) + AI (AI-generated) + application number, to eliminate potential cognitive confusion caused by the development of AI. The work of distinguishing the source during the authorization stage will directly impact the patent dissemination stage. Transparency in the source of generation can help market entities assess patent value, reduce transaction risks, and maintain a fair, just, and open market order.

In summary, during the machine-generated technical solution phase, AI service providers should be required to identify themselves to ensure that the generation process is traceable, so as to lower the risk of infringement and the cost of evidence. During the technical solution patent application phase, applicants should be required to disclose their use of AI to prevent damage to trust interests caused by information asymmetry. During the technical solution patent authorization phase, the National Intellectual Property Administration should take measures to guarantee the openness and transparency of patent sources, steadily promote the subsequent market dissemination of patents, and decrease patent transaction risks. Only by achieving information disclosure throughout the entire lifecycle from “source” to “downstream” can the dual needs of technological development and rights protection be balanced to lower the operational risks of the patent system.

6. Conclusions

The autonomous generation of inventions and creations by AI has become a fact. The law should serve social realities rather than remain static. Reforming the patent system will encourage sustainable innovation while creating a more equitable and inclusive technology ecosystem, where all actors, from large corporations to small startups, can participate and thrive [57]. Therefore, timely and appropriate adjustments to the patent system are highly aligned with China’s policy preferences to encourage the application and development of the AI industry. This paper primarily involves the investigation of the patentability of technical solutions generated by AI, patent examination standards, and issues related to patent ownership. The judicial practices and legal systems of certain countries and regions were compared and analyzed to identify suitable localized strategies for China. First, it was argued that AI-generated technical solutions possess patentability and that China should permit them to serve as the subject matter of patent rights under the premise of passing the “three criteria” review. Second, China should optimize the “three criteria” review standards for patents in accordance with the mechanistic characteristics of AI-generated solutions, given the inherent differences between AI-generated solutions and human creation. Third, China may consider establishing a parallel technical generation registration system for AI while maintaining the current inventor system. This approach would avoid legal and ethical issues and simultaneously adhere to the principle of good faith, contributing to clarifying the substantive contributions of AI to technical solutions. Fourth, the eligible entities for patent rights should be limited to the users and investors of AI regarding patent ownership issues. Contractual agreements should take precedence to ensure contractual freedom, and ownership should generally be attributed to the user in the absence of such agreements. Finally, China should optimize the operation of the patent system to enhance its efficiency and prevent potential infringement risks associated with AI. Moreover, a dedicated fast-track review and authorization channel should be established for AI-generated technical solutions to shorten the review period and extend the duration of the patent’s competitive advantage in the market. Given that AI-generated patents have cost advantages in terms of creation, their protection periods should be appropriately shortened to balance private and public interests and avoid the tragedy of the commons. With the purpose of enhancing the transparency of AI-generated technical solutions, corresponding information disclosure systems at each stage (from machine-generated technical solutions, patent applications, patent authorizations, to dissemination) should be established to reduce infringement risks and ensure the authority of information disclosure.

In summary, this paper provides a systematic overview of the patent challenges posed by AI-generated technology solutions, as well as a comprehensive response plan tailored to China’s patent system. As AI technology continues to advance and integrate into society, the legal challenges it poses must be identified and addressed at an early stage.