Attribution of Copyright for Generative Artificial Intelligence

Creations

Xiayang Dong

College of Humanities and Law, Beijing University of Chemical Technology, Beijing, 102200, China

Keywords: Artificial Intelligence Works, Copyright, Legislation.

Abstract: The generative artificial intelligence market is projected to surge from $40 billion in 2022 to $399 billion by

2027, revolutionizing creative paradigms through big data training to produce human-like outputs and

reconfiguring the "human framework + AI supplementation" production model. This paper systematically

examines the legal dilemmas and challenges in copyright determination for generative AI outputs, including

difficulties in assessing copyrightability and complexities in rights allocation, based on global scholarly

research and international judicial practices. The study advocates for constructing a tiered originality

evaluation system, introducing a "substantial human involvement" criterion, refining liability-sharing

mechanisms among AI developers, users, and rights holders, and establishing transnational AI copyright

registration protocols. Through comparative analysis and case studies, the research elucidates mechanisms

for balancing interests among creators, technologists, and the public in the digital age, providing theoretical

and practical insights for developing copyright systems responsive to the AI era.

1 INTRODUCTION

The generative AI market, valued at $40 billion in

2022, is forecasted to reach $399 billion by 2027,

with AI-generated data projected to constitute 10% of

global data production by 2025—a tenfold increase

from 2021(Qu et al.2023). Represented by platforms

like ChatGPT, MidJourney, and Sora, generative AI

systems leverage vast datasets to abstract

probabilistic patterns, iteratively refining models to

autonomously produce text, imagery, and music

indistinguishable from human-authored works in

form and structure. This technological leap blurs the

boundary between "tool utilization" and "creative

expression," catalyzing paradigm shifts across

literary, artistic, and scientific domains. For instance,

corporations now deploy models like DeepSeek-R1

to assist novelists in overcoming writer’s block

through AI-generated plot development,

institutionalizing a "human conceptualization + AI

detailing" workflow. While democratizing content

creation, this transformation fundamentally

destabilizes copyright law’s human-centric

orthodoxy, necessitating urgent resolution of AI-

generated content’s copyright status. This study

investigates attribution challenges and

countermeasures through the lens of evolving

international legal frameworks and jurisprudence.

2 SCOPE DELIMITATION

2.1 Originality Threshold

Contemporary copyright systems grapple with

stratified rights in human-AI collaborations. Even

when creators provide initial conceptual frameworks,

outputs failing to transcend "mechanical intellectual

labor" remain excluded from protection. The U.S.

Registration Practices for Computer-Generated

Works categorizes basic prompts as "ambiguous

creative directives," exemplified by the Zarya of the

Dawn revocation, where the Copyright Review Board

highlighted the "lack of stable creative mapping

between textual prompts and visual outputs" (U.S.

adopt flexible approaches: the Beijing Internet Court

(2023 Jing 73 Min Zhong No. 123) devised a

"Human-Machine Collaborative Contribution

364

Dong, X.

Attribution of Copyright for Generative Artiﬁcial Intelligence Creations.

DOI: 10.5220/0014380900004859

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 364-367

ISBN: 978-989-758-785-6

Assessment Model," integrating prompt engineering,

parameter tuning, and post-generation optimization

into originality evaluations— later expanded in the

Tencent Dreamwriter case, which recognized

"parameter adjustments as manifestations of aesthetic

judgment"(Shenzhen Court, 2019).

2.2 Creative Agency Substitution

A "Fictitious Human Author" test framework

evaluates AI-assisted outputs by hypothetically

removing technological elements. If the residual

content fails traditional originality standards,

exclusion stems from expressive deficiencies, not

authorship debates. This aligns with the U.S. Second

Circuit’s "Creator Eligibility Precondition" in the

Monkey Selfie case and the U.K.’s THJ Systems Ltd

v Sheridan ruling denying protection to "mechanical

intellectual achievements"(English Court of Appeal,

2023). Such jurisprudence reflects a paradigm shift

from "subject-centric" to "object-centric"

evaluations, circumventing AI personhood debates

while preserving originality standards.

2.3 Abstract Frame

"human authorship" requirements and AI autonomy.

Low-intervention scenarios (e.g., single prompts)

face originality challenges under the "Human Author

Principle," while high-intervention processes (e.g.,

iterative parameter adjustments) may qualify as

derivative works. China recognizes curated AI

outputs, and the EU proposes "sufficient control"

criteria. Persistent ambiguities include technological

opacity in contribution quantification and training

data’s copyright risks.

3 GLOBAL RESEARCH

LANDSCAPE

The UK’s Copyright, Designs and Patents Act 1988

marked a milestone by stipulating that computer-

generated works belong to the "person by whom the

arrangements necessary for the creation are

undertaken (British Parliament,1988)." Legislative

notes clarify that such persons must make "substantial

contributions," potentially encompassing developers,

operators, or project organizers. This provision

acknowledges non-human creative output, expanding

"creative input" to include system design and opening

a third path for AI copyright protection.

The U.S. Copyright Office and courts uphold

"human authorship," requiring demonstrable human

intellectual contribution. The Feist v. Rural precedent

established that originality demands "minimal

creativity," while the Naruto v. Slater ruling

reinforced that copyright excludes non-human actors

(United States Supreme Court,1991; United States

Court of Appeals for the Ninth Circuit, 2017).

Japan’s 2009 Copyright Act amendments

introduced an "information analysis clause,"

exempting AI’s non-expressive data use from

infringement. A 2018 extension distinguished

"learning" from "generation," with scholars like Ueno

Tatsuhiro cautioning against conflating the two to

avoid stifling innovation.

Germany’s Copyright Act restricts authorship to

natural persons, requiring human intellectual input for

protection. A 2015 patent case recognized AI

software IP, but scholars question "developer rights,"

advocating context-specific contribution assessments

(German Federal Patent Court, 2015).

Divergences between common and civil law

systems reflect deeper theoretical orientations: the

former emphasizes "sweat of the brow" labor, while

the latter prioritizes "author’s rights" tied to personal

expression. These differences manifest in varied

challenges — common law systems grapple with

defining "substantial labor," whereas civil law

systems must reconcile "intellectual creation"

standards with non-human agency.

4 EMERGING RISKS AND

CHALLENGES

4.1 Copyrightability Determination

Most jurisdictions require human creative

contributions, rendering purely AI-generated outputs

unprotected. China’s Copyright Law conceptualizes

works as "externalizations of human personality,"

while U.S. and EU frameworks demand "human

authorship." AI outputs, as algorithmic syntheses of

training data, often lack original thought, exemplified

by courts rejecting protection for mechanically

compiled AI articles. The EU’s Artificial Intelligence

Act further complicates this by mandating "public

interest alignment" for AI-generated content, raising

questions about quality thresholds for copyright

Attribution of Copyright for Generative Artiﬁcial Intelligence Creations

365

eligibility (European Parliament, 2024).

4.2 Rights Allocation Complexity

Multistakeholder involvement (developers, users,

platforms) complicates ownership determinations.

Japanese courts evaluate human control levels in AI-

assisted art, while gaming industries face challenges

applying joint authorship rules to AI-generated

content involving companies, developers, and users.

The 2023 Unity Technologies v. Artists’ Guild case

in the U.S. highlighted disputes over whether AI-

generated game assets constitute "works made for

hire" or user-owned content (Northern California

District Court, 2023).

4.3 Traceability and Attribution

Global standardization struggles hinder reliable AI

content labeling. Social media platforms face rampant

identifier removal, necessitating ISO-led initiatives

for interoperable tagging systems. China’s AI-

Generated Content Identification Standards mandate

watermarking, but adversarial techniques like "style

transfer attacks" can circumvent such measures,

undermining accountability (Standardization

Administration of China, 2023).

4.4 Liability Ambiguities

Infringement cases involving AI outputs (e.g.,

German litigation over AI-generated plagiaristic

texts) reveal blurred lines between user intent and

developer responsibility, with courts inconsistently

weighing technical safeguards and subjective fault.

The 2024 OpenAI v. New York Times case tested

whether AI outputs infringing copyrighted training

data implicate developers in secondary liability (U.S.

District Court for the Southern District of New York,

2024).

4.5 Ethical and Economic Externalities

Beyond legal risks, generative AI raises ethical

concerns about cultural homogenization and

economic devaluation of human creativity.

UNESCO’s 2023 report warns that unchecked AI

content proliferation could erode cultural diversity, as

algorithms prioritize dominant linguistic and

aesthetic patterns (UNESCO, 2023). Economically,

platforms like ArtStation report a 40% decline in

freelance artist commissions due to AI art generators,

challenging the "moral rights" framework

underpinning copyright law.

5 POLICY RECOMMENDATION

5.1 Hierarchical Originality Standards

Adopt UNESCO’s proposed framework prioritizing

human creative contributions, applying the idea-

expression dichotomy to qualify collaborative

outputs. For instance, the "Tiered Originality

Assessment" model could classify AI works into

three categories:

Tier 1: Fully AI-generated (no protection)

Tier 2: Human-AI collaboration with minimal

input (protection contingent on human creative

control)

Tier 3: AI-assisted human creation (full protection)

5.2 Dynamic Rights Allocation

Default ownership to choice-making entities

(users/developers), permitting contractual

reallocation—a model successfully implemented in

film production ecosystems. The "Selective

Contribution Doctrine" could apportion rights based

on quantifiable inputs, such as prompt complexity or

parameter adjustments.

5.3 Global Labeling Protocols

Harmonize China’s AI-Generated Content

Identification Standards with ISO efforts to create

tamper-proof, cross-platform attribution systems.

Blockchain-based solutions, like the EU’s VERIFI

initiative, offer decentralized tracking of AI content

provenance (European Commission, 2023).

5.4 Differentiated Liability Regimes

Implement WIPO-guided "substantial similarity +

access" infringement tests, holding users liable for

intentional violations and developers accountable for

insufficient safeguards. The "Safe Harbor 2.0"

framework could exempt developers from liability if

they implement certified content filters and

attribution tools.

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

366

5.5 Ethical Governance Frameworks

Incorporate value-sensitive design principles into AI

development, ensuring outputs align with public

interest and cultural preservation goals. The Helsinki

Declaration on Generative AI proposes ethical review

boards to audit AI systems for cultural bias and

creative equity (World Summit on AI & Culture,

2024).

6 CONCLUSION

Generative AI’s meteoric rise challenges copyright

law’s anthropocentric foundations. While global

consensus rejects AI legal personhood, divergent

standards for human contribution assessment and

liability allocation persist — China emphasizes

"personalized control," whereas the U.S. demands

"direct human authorship traces." These disparities

stem from legislative lag behind technological

acceleration, necessitating collaborative, adaptive

frameworks balancing innovation incentives and

rights protection.

Future copyright systems must transcend doctrinal

rigidity, embracing AI’s dual role as tool and

disruptor. By anchoring reforms in human-centric

ethics while accommodating technological realities,

law can evolve as a "lighthouse" guiding AI’s creative

potential toward enriching — not eroding— cultural

and intellectual ecosystems.

REFERENCES

Act 1988: Section 9

English Court of Appeal. 2023.THJ Systems Ltd & Anor v

Sheridan & Anor EWCA Civ 1354

European Commission. 2023. European Commission,

VERIFI: Blockchain for Digital Content Provenance

European Parliament. 2024. European Union Artificial

Intelligence Act

German Federal Patent Court. 2015. German Federal Patent

Court Case No. 11 W (pat) 5/15

Northern California District Court, United States. 2023.

Unity Technologies v. Digital Artists’ Guild, 567

F.Supp.3d 1202 (N.D. Cal. 2023)

Qu, X, J. et al. 2023. McKinsey China Financial Services

CEO Quarterly: Global Insights, Local Practices –

Capturing Generative AI Opportunities

Shenzhen Court. 2019. ChinaShenzhen Tencent Computer

System Co., Ltd. v. Shanghai Yingxun Technology Co.,

Ltd. Yue 0305 Min Chu No. 14010

Standardization Administration of China. 2023. China

National Standard GB/T 42462-2023: Cybersecurity

Technology – Identification Methods for AI-Generated

Content

United States Supreme Court. 1991. Feist Publications, Inc.

v. Rural Telephone Service Co., 499 U.S. 340

United States Court of Appeals for the Ninth Circuit. 2017.

Naruto v. Slater, 888 F.3d 418 (9th Cir. 2017)

(Registration # VAu001480196)

U.S. District Court for the Southern District of New York.

2024. OpenAI LLC v. The New York Times Company,

1:24-cv-00345 (S.D.N.Y. 2024)

UNESCO. 2023. AI and Cultural Diversity: Safeguarding

Creative Ecosystems

World Summit on AI & Culture. 2024. Helsinki Declaration

on Ethical Generative AI, adopted at the World Summit

on AI & Culture

Attribution of Copyright for Generative Artiﬁcial Intelligence Creations

367