On the Ownership of the Rights of Artificial Intelligence-Generated

Works: Systematic Response Based on the Identification and Norms

of Legal Subjects

Yaze Li

Civil and Commercial Law School, Southwest University of Political Science and Law, Chongqing, 401120, China

Keywords: Generative Artificial Intelligence, Copyright Law, Legal Subject Attribute Positioning, Gradient Weight.

Abstract: The essence of the problem of copyright ownership of audiovisual works generated by artificial intelligence

is that in the digital age, there is a crisis in the identification and regulation of legal subjects. Traditional civil

and commercial law constructs the framework of rights allocation based on ‘anthropocentrism’, which leads

to structural conflicts in new creative features such as algorithm autonomy and data resource dependence.

Through the analysis of the legal subject, this paper analyzes the institutional problems faced by the content

of artificial intelligence generation in the qualification of the right subject, the division of contribution weight

and the attribution of responsibility, and puts forward the construction of ‘dynamic right generation model’

to reshape the civil and commercial law system. Through the comparison method and case analysis method,

the feasibility of the ‘three-subject framework’ and the ‘contribution chain distribution mechanism’ is

demonstrated, which provides a theoretical path to fill the gap between technological innovation and legal lag.

1 INTRODUCTION

When the deep learning algorithm gradually produces

the ability to ‘think’ in the Internet age, and the

generative artificial intelligence breaks through the

boundary of human imagination, the production order

of the cultural industry faces a new crisis with the

development of science and technology. In the

collision between this technological revolution and

the legal tradition, a fundamental question becomes

clearer: when the autonomy of the algorithm

dissolves the physical carrier of ‘creative intention’,

when the data dependence separates the causal chain

of ‘author-work’, is the existing copyright system

with anthropocentrism as the theme experiencing an

unprecedented paradigm crisis in the history of

civilization?

The current legal system has exposed deep

structural contradictions in dealing with generative

artificial intelligence. Although judicial practice tries

to maintain the logical self-consistency of

‘instrumentalism’ and attributes the ownership of

rights to the degree of intervention of human

operators, it is difficult to explain the autonomous

evolution of creative decisions generated by artificial

intelligence in the black box of algorithms. When the

generated content is neither the technical

achievement preset by the developer nor the

intellectual expression completely controlled by the

user, the subject identification logic of the traditional

civil and commercial law ‘either-or’ faces the lack of

judicial interpretation. It is worth pondering whether

the ‘machine reading’ of massive works in the

process of data training constitutes a digestion of the

creative intention of human creators, and whether the

unexpected ideas generated by algorithm iteration can

be included in the scope of existing rights generation.

The academic community’s response to this

shows two different propositions of theoretical

construction and institutional construction. Some

scholars advocate the expansion of the subjectivity of

the existing law, trying to fill the vacancy of the

system through the creation of legal personality, but

they are faced with the contradiction between the

compatibility with the traditional civil law value

system; Another part of scholars tried to construct a

new contribution evaluation model by improving the

distribution of rights and interests between different

subjects, which was also plagued by the problem of

how to quantify the black box of technology (Wu,

2020 & Geiger, 2024). These explorations reveal the

complexity of the problem at different levels, but

Li, Y.

On the Ownership of the Rights of Artiﬁcial Intelligence-Generated Works: Systematic Response Based on the Identiﬁcation and Norms of Legal Subjects.

DOI: 10.5220/0014359700004859

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 1st International Conference on Politics, Law, and Social Science (ICPLSS 2025), pages 215-220

ISBN: 978-989-758-785-6

215

have not yet constructed an effective legal normative

framework that is deeply compatible with the creative

characteristics of generative artificial intelligence.

When algorithmic autonomy has become a prominent

feature in the creative process, should institutional

innovation adhere to the bottom line of human-

centeredness, or need to breed a new type of rights

ecology adapted to human-machine symbiosis?

Based on the paradigm crisis brought by

generative artificial intelligence to the legal system at

this stage, this paper focuses on the innovation of the

existing laws on the regulation of subjectivity, and

tries to put forward solutions to the three dilemmas

faced by the current laws: First, to solve the

contradiction between the ‘tool-subject’ binary

opposition, and to create a normative space for the

autonomy of the algorithm in the civil and

commercial legal system; Secondly, in the

transmission chain of ‘data-algorithm-output’, a

traceable and quantifiable contribution identification

system is constructed. Third, balance technological

innovation incentives, protect fairness and justice in

the distribution of rights and interests, and prevent

technological advantages from evolving into a

monopoly of rights. This paper hopes to open up a

new institutional channel for man-machine

collaborative creation by creating a ‘dynamic rights

generation model’ and a ‘gradient weight evaluation

mechanism’, so that the copyright law will still

radiate new vitality in the era of artificial intelligence.

2 DECONSTRUCTION OF THE

PARADIGM OF LEGAL

SUBJECTIVITY CRISIS

As the proportion of algorithmic content in the digital

content market exceeds 60 %, new challenges facing

the legal system have emerged, Under the

background that the algorithm system gradually

breaks through the boundary of tools and shows its

characteristics of quasi-subjectivity, does the subject

category of the current ‘copyright law’ with natural

persons and legal persons as the core have an

institutional conflict with the gradually developing

technological civilization (WIPO, 2024)? The

traditional civil and commercial law theory has fallen

into the dual dilemma of " absence of subjectivity "

and " imbalance of interest distribution " when

dealing with the ownership dispute of generative

artificial intelligence (Coase, 1960). It is urgent to

realize the paradigm transformation through the

expansion of legal subjectivity theory and

institutional innovation.

At present, judicial practice always adheres to the

logic of ‘instrumentalism’ and regards artificial

intelligence as the dominant creative medium. The U.

S. Copyright Office’s 2023 ‘Generative Artificial

Intelligence Copyright Policy Statement’: ‘Human

Creative Control’ is the core standard for rights

attribution (United States Copyright Office, 2023). In

the typical case of generative artificial intelligence,

‘Fehling Case’, Fehling Law Firm uses legal data

analysis software to automatically generate ‘judicial

big data analysis report of film and television

entertainment industry’, which covers data screening,

statistics and visual chart design. However, the court

held that the core content of the report was

automatically completed by the algorithm and lacked

the ‘original expression of natural persons’, which did

not constitute a work in the sense of copyright law. In

spite of this, the court still recognized the user’s

investment in the operation of the software, giving it

the right to prohibit unauthorized dissemination by

others (Beijing Internet Court, 2018). This judgment

reflects the current legal attitude towards generative

artificial intelligence. It still focuses on the traditional

" tool theory, " regards generative artificial

intelligence as a creative tool, and regards whether the

user’s intervention in the generation process of

artificial intelligence is sufficient to constitute

original labor as the criterion for judgment. However,

when the generated content exceeds the developer’s

preset range, the causal relationship between the

user’s instructions and the output results is also

separated due to the autonomy of the algorithm. The

developer is not directly involved in the content

generation, and the user cannot fully control the

generated results. At this time, the ‘tool theory’

ownership standard has fallen into a theoretical

dilemma. For example, the image created by deep

forgery technology with adversarial training is not the

design result obtained by the developer directly, nor

the creation result that the user can fully control.

At the same time, the limitations of the ‘natural

person-legal person’ dual subject structure of

traditional civil and commercial law have become

more and more significant in the context of the

gradual enhancement of the autonomy of modern

algorithms. On the one hand, the behavioral

responsibility of AI-generated content is difficult to

clearly identified, and the boundary of responsibility

between developers, users and AI systems is blurred

due to the black box of technology; on the other hand,

the mechanism of income distribution also faces the

risk of imbalance. It is difficult to quantify and

ICPLSS 2025 - International Conference on Politics, Law, and Social Science

216

objectively measure the basic contribution of

developers to the algorithm architecture and the

proportion of rights and interests invested by users.

The more fundamental contradiction is that the

autonomy of the algorithm leads to the opacity of the

creative process, which leads to the contradiction in

the logic of ‘creative intention-expression result’ that

the traditional copyright law relies on. The ‘Artificial

Intelligence Act’ promulgated by the EU attempts to

strengthen the responsibility of developers through

technical transparency, but it does not respond to the

criteria of " developer presupposition " and " user

control " that cannot prevent algorithms from

generating autonomous creation and their autonomy

(Artificial Intelligence Act, 2024). If Coase’s

property rights theory is introduced into the field of

artificial intelligence generation, the initial rights

allocation will take into account both efficiency and

fairness (Coase, 1991). Users can achieve deep

intervention in artificial intelligence by adjusting

parameters, just like the labor contribution in the

‘partnership enterprise law’, giving them copyright

property rights. The developer’s basic contribution to

the algorithm architecture can rely on the ‘technical

contribution weight’ to obtain income distribution.

This path not only retains the tradition of legal

protection of creative labor, but also reserves the

institutional interface for the autonomous

development of future algorithms.

3 STANDARD PERSPECTIVE ON

THE DILEMMA OF RIGHTS

OWNERSHIP

In the scenario generated by artificial intelligence, the

presupposition of the concept of ‘human author’. On

the one hand, if the generated content is completely

free from human intervention, it will fall into a state

of ‘right vacuum’. The current legal system has not

yet formed a unified normative framework for the

ownership of the rights of artificial intelligence-

generated content, and there are two main

contradictions between legislative gaps and

interpretation disputes. Article 3 of China’s

‘Copyright Law’ regards ‘natural person creation’ as

the core element of the identification of works, but it

is not clear whether the content generated by artificial

intelligence can be included in the category of

creation. Although the EU’s ‘Artificial Intelligence

Act’ requires developers to take responsibility for

content compliance, it does not break through the

logic of ‘anthropocentrism’ (Artificial Intelligence

Act, 2024).

At the same time, at the level of normative

interpretation, the definition of the qualification of

‘author’ in judicial practice is not yet clear. The

Beijing Internet Court regarded artificial intelligence

as a ‘reative tool’ in the ‘Feelin case’, and determined

that the user’s original expression formed by

parameter adjustment can become the subject of

rights (Beijing Internet Court, 2018). Then when

artificial intelligence has substantial intellectual input

to content generation, it faces another problem,

whether developers and users can constitute co-

authors. Such contradictions reflect the conflict

between the principle of ‘the uniqueness of the

creative subject’ in the traditional copyright law and

the continuous development of the characteristics of

artificial intelligence technology at the present stage.

At present, the difficulties in the application of

law mainly focus on the technical aspect of the right

division mechanism. The rules of cooperative works

and commissioned works in the current law are

premised on the traceability of human behavior, but

the black box of the algorithm of artificial

intelligence-generated content makes it difficult to

quantify the contribution. Although Article 7 of the

‘Measures for the Administration of Generative

Artificial Intelligence Services’ requires the

legalization of data sources, it does not solve the

relationship between the ownership of the trained data

and the ownership of the generated data. This kind of

normative lag makes the judicial practice rely heavily

on the contract agreement but the existing but user

agreement rights transfer clause often has the risk of

obvious unfairness.

The dispute over the legality of data training has

exacerbated the problem of ownership. During data

training, massive works are faced with the problem of

defining rights conflicts caused by ‘machine reading’.

Wu Handong advocates the rational use of rules in

data mining (Wu, 2020). Japanese policy documents

show that the industry is deeply worried about data

ownership and needs to establish a transparent data

tracking system (Japan's Cabinet Office, 2024). In

judicial practice, data mining, like artificial

intelligence, is used as a creative tool, but training

data with copyright protection may bring new

infringement risks. The contradiction between this

normative gap and technological progress has caused

the dilemma of the ownership of the rights of artificial

intelligence to generate audiovisual works.

On the Ownership of the Rights of Artiﬁcial Intelligence-Generated Works: Systematic Response Based on the Identiﬁcation and Norms of

Legal Subjects

217

4 ANALYSIS OF THE DILEMMA

OF THE OWNERSHIP OF

ARTIFICIAL

INTELLIGENCE-GENERATED

WORKS

The identification of the ownership of artificial

intelligence-generated works is special. The existing

legal framework based on the copyright ownership

rules preset by human creators (such as the authorship

identification standard stipulated in Article 11 of

China’s ‘Copyright Law’) is difficult to apply directly.

However, there are still logical faults and institutional

gaps in the theory of ‘AI tools’ and ‘fiction authors’

proposed by the academic community in terms of

ownership distribution standards and incentive

mechanism design (Geiger, 2024). The current

in dealing with the characteristics of artificial

intelligence creation. First of all, the subject of rights

system still uses the single mode of " human or legal

person, " which obviously cannot fit the new form of

mixed creation of human-computer cooperation. In

addition, the definition of " subject " generated by

artificial intelligence is vague, which makes the

legitimacy of data mining depend on case discretion,

and there is no expected definition standard

(Floridi, 2023).

The rapid development of industrial practice is

pushing forward the change of law. Japanese policy

documents point out that the application of generative

artificial intelligence in the fields of film and

television soundtracks and virtual actors has led to the

emergence of the phenomenon of ‘virtual creation

subject’. The human-led principle established by

China’s ‘Fehling case’ may encounter fundamental

challenges in the context of the increasing proportion

of AI-generated content. The EU has implemented a

risk classification system, based on which a

hierarchical regulatory framework has been

constructed, and some progress has been made in

security management (Japan's Cabinet Office, 2024).

However, the system does not touch on the core issue

of rights attribution. This gap between institutional

supply and technological innovation urgently needs

to be integrated and bridged by theory.

To break through the shackles of the traditional

‘instrumentalism’, it is necessary to implant the

system design of ‘class subject’ in the civil and

commercial law system. The EU’s ‘Artificial

Intelligence Act’ assigns responsibility to ‘digital

legal persons’ for high-risk systems, and the Japanese

policy document proposes to treat developers as ‘co-

authors’ (Artificial Intelligence Act, 2024 & Japan's

Cabinet Office, 2024). These are attempts to expand

the subjectivity of the law.

5 EXPLORING THE

RECONSTRUCTION PATH OF

LEGAL SUBJECTIVITY

THEORY

Wu Handong proposed to give artificial intelligence

a ‘limited legal personality’, making it a legal

relationship participant with specific legal capacity

(Wu, 2020). In EU legislation, this idea has not yet

been fully formed. The ‘Artificial Intelligence Act’ of

EU legislation is reflected in the establishment of

‘digital legal person’ responsibilities for high-risk

artificial intelligence systems, but has not yet been

extended to the field of copyright (Artificial

Intelligence Act, 2024). The Japanese policy suggests

that developers should be regarded as ‘co-authors’,

and breakthroughs should be made in the existing

main framework. This progressive improvement

method is practical and operational (Japan's Cabinet

Office, 2024).

5.1 Imbalance in the Allocation of

Rights in the Current User

Agreement and Countermeasures

Most of the current user agreements have imbalances

in the allocation of rights. Taking the Mid journey

service clause as an example, the agreement of

‘copyright ownership platform of products’ conflicts

with the invalidity rule of the standard clause of

Article 497 of the Civil Code of China. This article

advocates drawing on the ‘author presumption rule’

of Article 32c of the German ‘Copyright Law’ (Hilty

& Köklü et al., 2021). The platform should bear the

burden of proof, otherwise it should regard the user

as the subject of rights. In terms of responsibility

distribution, a three-tier framework system should be

constructed: that is, the three-tier responsibility

system of ‘user-developer-platform’. Among them,

the user’s actual control over the generated content

will define its rights boundary. Developers bear the

strict responsibility of systemic risk because of the

defects of the algorithm. The platform shall perform

the corresponding duty of care in accordance with the

provisions of Article 42 of the ‘E-commerce Law’,

and build a content traceability mechanism and

infringement warning system. When users know that

the data is flawed but still use the resulting content,

ICPLSS 2025 - International Conference on Politics, Law, and Social Science

218

they need to bear the corresponding liability

according to Article 52 of the Copyright Law. If the

platform fails to fulfill its duty of care (such as not

building a traceability mechanism), it should bear

supplementary responsibility. Developers suffer

from systematic infringement due to defective

algorithm design, and need to bear serious

responsibility.

5.2 Institutional Innovation Path of

Responsibility Distribution and

Interest Balance Mechanism

In the process of promoting the innovation of the

interest balance mechanism, the system design can be

carried out along three paths: One is to refer to the

EU’s special protection model for databases and

establish neighboring rights for AI-generated content

(Artificial Intelligence Act, 2024). Secondly, Japan’s

‘algorithm contribution value evaluation’ technology

can be introduced to construct a quantitative rights

allocation system (Japan's Cabinet Office, 2024).

Thirdly, the application scope of statutory license

should be broadened and a set of perfect training data

authorization mechanism should be constructed.

These schemes are not only connected with the

traditional system, but also meet the special needs of

technological development.

More fundamentally, the reconstruction of the

theory of legal subjectivity needs to adapt to the value

system of civil and commercial law. When the

content generated by the algorithm has the nature of

social welfare, the scope of statutory license can be

appropriately broadened according to the provisions

of Article 185 of the Civil Code. If it comes to public

areas such as cultural communication, the fair use

mechanism of Article 24 of the Copyright Law should

be opened. This layered response strategy not only

ensures that the traditional system is not subverted,

but also leaves room for technical iteration, and

finally achieves the dual legislative purpose of

encouraging innovation and maintaining fairness.

6 CONCLUSION

The research status of artificial intelligence law in

China shows an imbalance of ‘structure-function’: in

terms of theoretical construction, most of the

subjective research is limited to philosophical

speculation, and does not connect with the systematic

development of civil and commercial law system. In

the study of comparative law, the reference of EU and

Japan’s legislation is limited to the transplantation of

corresponding provisions, and does not deeply

analyze the social reality contradictions generated by

the system. In terms of practical response, judicial

decisions have not yet established stable case-like

rules, and many policy recommendations remain at

the level of principle declaration.

Therefore, future research needs to focus on

constructing a binary subject of ‘human-AI’ and

designing more detailed rights identification

standards. The legal system can distinguish the

ownership of rights according to the control intensity

of human data screening, parameter adjustment and

result optimization by constructing the index system

of ‘creative participation’. At the same time, it is

necessary to improve the evaluation framework of

data training legitimacy, and clarify the difference

and legal nature between basic model training and

fine-tuning optimization. More importantly, it is

essential to explore the right distribution mechanism

of ‘creative contribution chain’, and achieve the

dynamic balance of interest distribution through more

perfect and reasonable technology. Through the

continuous optimization of the existing legal system,

this systematic reconstruction can not only maintain

the core function of the existing copyright law to

stimulate innovation, but also adapt to the new mode

of human-machine collaborative development in the

future.

REFERENCES

Artificial Intelligence Act (European Union Regulation No.

2024/1689, 2024).

Beijing Internet Court, (2018) Beijing 0491 Civil Judgment

No. 239.

https://wenshu.court.gov.cn/website/wenshu/181107A

NFZ0BXSK4/index.html (Accessed: April 26, 2019)

C. Geiger, When the robots (try to) take over: Of artificial

intelligence, authors, creativity and copyright

protection. In: F. Thouvenin, A. Peukert, T. Jaeger, C.

Geiger (Eds.), Kreation innovation märkte - creation

innovation markets (Springer, Berlin, 2024).

H. Wu, Questions on copyright law of works generated by

artificial intelligence. Chin. Law (3) (2020).

Japan's Cabinet Office, Know-how property advancement

plan 2025. (2024).

L. Floridi, The ethics of artificial intelligence: Principles,

challenges, and opportunities (Oxford University Press,

2023).

R.H. Coase, The problem of social cost. In: C.

Gopalakrishnan (Ed.), Classic papers in natural

resource economics (Palgrave Macmillan, London,

1960).

On the Ownership of the Rights of Artiﬁcial Intelligence-Generated Works: Systematic Response Based on the Identiﬁcation and Norms of

Legal Subjects

219

R.H. Coase, The nature of the firm (Oxford University

Press, 1991).

R.M. Hilty, K. Köklü, V. Moscon et al., International

instrument on permitted uses in copyright law. IIC 52,

62-67 (2021).

United States Copyright Office, Compendium of U.S.

World Intellectual Property Organization, Global

innovation index 2024: Unlocking the promise of social

entrepreneurship (WIPO, Geneva, 2024).

ICPLSS 2025 - International Conference on Politics, Law, and Social Science

220