Study on Copyright Infringement Liability of Short Video Platforms

from the Perspective of Intelligent Algorithm Recommendation

Xiaopin Lyu

and Yaqing Yang

2,*

Faculty of Law, Shandong Agricultural University, Tai'an, Shandong, 271018, China

Faculty of Law, Zhejiang University of Technology, Hangzhou, Zhejiang, 310014, China

Keywords: Intelligent Algorithm Recommendation Technology, Short Video Platform, Copyright, Principle of

Technology Neutrality.

Abstract: In the context of the Internet era, with the continuous development of the short video industry, the intelligent

algorithmic recommendation technology to provide personalized content for users is widely used by short

video platforms. However, the use of this technology has facilitated the occurrence of infringement on a large

scale, resulting in damage to the rights and interests of copyright owners. China's existing legislation lacks

specific and detailed provisions on the specific liability scenarios of "algorithmic recommendation". In

judicial practice, there is no uniform standard for determining whether and how the platform should be held

liable, and the academic community also has different opinions. To address this issue, it is necessary to

analyze the issue from the perspective of the traditional safe haven principle and in conjunction with specific

theories such as the principle of technological neutrality. In addition, it is also necessary to explore the

divergence in judicial practice on the duty of care of the platform in the light of classic judicial cases. On this

basis, further specific paths for reconfiguring the boundaries of responsibility are proposed.

1 INTRODUCTION

As the market scale of the short video industry is

growing explosively and the group of short video

creators is getting bigger and bigger, the intelligent

algorithm recommendation technology has become

the core means for platforms to enhance user

stickiness and commercial revenue. Intelligent

algorithm technology can efficiently collect, mine

and parse the massive user behavior data accumulated

on the platform, so as to further accurately identify

the interests and preferences of each user. Ultimately,

through information matching technology, the

content that is highly suitable for the personalized

needs of the user is automatically recommended to

the user's field of vision, so as to realize the precise

and automated information push service (Le &Wang

et al., 2025).

However, this technology has also contributed to

the large-scale dissemination of infringing behaviors

such as editing and handling of movie and television

works, resulting in frequent damage to the interests of

Corresponding author

right holders and a sharp increase in the cost of social

justice. In this context, the traditional "safe harbor

principle" is lagging behind due to its reliance on the

"notice-and-delete" rule, and platforms often use

"technology neutrality" as a reason to avoid

responsibility. At the same time, the existing

technical difficulties are the high cost of infringement

identification due to massive short video content, and

the algorithm accuracy and efficiency still need to be

improved. The existing law lacks detailed provisions

on the specific liability scenarios of "algorithmic

recommendation", and the courts are divided on the

determination of the platform's duty of care in judicial

practice.

In this context, in view of the current legislative

status, judicial practice differences and theoretical

disputes, this paper is to summarize the existing focus

of controversy from the specific judicial practice -

whether the platform is responsible for the results of

the algorithmic recommendation, and how to be

responsible for. From the specific theory, to put

forward a specific path of reconstruction of the

boundary of responsibility.

Lyu, X. and Yang, Y.

Study on Copyright Infringement Liability of Short Video Platforms from the Perspective of Intelligent Algorithm Recommendation.

DOI: 10.5220/0014359300004859

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 197-203

ISBN: 978-989-758-785-6

197

2 RESEARCH STATUS

2.1 Status of Chinese legislation

China's current legal framework has gradually

improved the regulation of platform liability for

algorithmic recommendations, but there are still gaps

in the system's articulation. The Copyright Law, as

amended in 2021, expands the scope of copyright

protection by introducing the concept of "audiovisual

works", but fails to clearly define the liability of

platforms in the context of algorithmic

recommendations. Articles 1194-1197 of the Civil

Code, which are the basic norms of network

infringement, establish the joint and several liability

of platforms when they "know or should know" of the

infringement, and although it stipulates the obligation

of "notification and deletion", the scope of

application is still limited to the passive response

mechanism, which fails to cover the new

communication form of algorithmic active

recommendation. The State Council issued the

"Regulations on the Protection of the Right to

Network Dissemination of Information" to continue

the principle of safe harbor, requiring platforms to

passively respond to requests for deletion, and

strengthening the platform's obligation to dispose of

the aftermath, but did not make the mandatory

requirements for platforms to review and filter video

content. The "Regulations on the Administration of

Algorithmic Recommendation of Internet

Information Services" came into force in 2022. These

regulations added some important new rules. They

clearly state that algorithmic recommendation service

providers must not use technical means to spread

infringing content. Providers must set up a system to

handle user complaints. They should also keep

improving their algorithms to reduce the risk of

infringement. In addition, they must take active steps

to limit the spread of illegal content, such as deleting

it and controlling its visibility.

However, China's existing laws lack detailed

provisions on the specific liability scenarios of

"algorithmic recommendation". It is worthwhile to

further study how to apply the relevant provisions in

specific cases, and to clarify the responsibility and

duty of care borne by the platform.

2.2 China's Judicial Status and Policy

Exploration

At the judicial level, some of the jurisprudence

represented by the case of iQIYI v. a short-video

platform adopted the "red-flag standard". They found

that the platform "should have known" about the

infringing content of the hit series. They broke the

boundary of the traditional "notice-and-delete" rule.

It reflected an expansive judicial interpretation of the

duty of care in the context of technological innovation

(Beijing Haidian District People's Court, 2018). In

terms of industry governance, although the "Internet

Information Service Algorithm Recommendation

Management Regulations" advocates the value

orientation of algorithms "upward and good", the

design of specific provisions of copyright protection

is still too rough, and there is a lack of clear and

operable specific implementation rules.

Therefore, some scholars suggest drawing on the

EU Digital Single Market Copyright Directive to

promote the adoption of filtering technologies such as

Content ID by platforms in order to build a preventive

liability system. However, other scholars oppose the

introduction of the mandatory filtering mechanism in

the EU Digital Single Market Copyright Directive,

arguing that it may inhibit innovation, emphasizing

the principle of technological neutrality and the

principle of migratory platforms to determine

liability. But the Chinese legislation has not yet

included filtering technology in the scope of

mandatory obligations (Tang, 2017 … Li, 2025).

Currently, legislation is lagging behind

technological development, resulting in the criteria

for determining platform liability oscillating between

judicial discretion and technological circumvention.

In judicial practice, the court has not yet formed a

unified standard for the judgment of the duty of care

of the platform under the algorithmic

recommendation scenario, and there is disagreement

between the theoretical community and the practical

community on the delineation of the boundary of

responsibility (Sin, 2024). The principle of

"technological neutrality" has also been gradually

exposed as a loophole for abuse in the process of

application, and some platforms have utilized it to

circumvent the obligation of active review, which

reflects the risk of disorder in the rule system (Peng

& Ding, 2022). How to establish a mechanism to

balance technological innovation and rights

protection in order to regulate the short video

platform algorithmic recommendation infringement

phenomenon has become an important issue that the

to.

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

198

3 THEORETICAL

CONTROVERSIES AND

ADJUDICATIVE

DISAGREEMENTS

In judicial practice, platforms often rely on the "safe

haven principle" to launch the defense, to receive

notice has been taken down, blocking the infringing

content as a defense, so that the traditional "notice-

deletion" mechanism is in a predicament, the

phenomenon of infringement is worsening

(Peng &

Ding, 2022). At the same time, in the context of the

increasing popularity of algorithmic recommendation

technology, platforms advocate "technological

neutrality" as a means of avoiding the obligation to

review algorithmically recommended content, as well

as the responsibility arising therefrom. This

phenomenon has led to differences in academic and

judicial practice on the determination and division of

the duty of care assumed by the platform.

3.1 Dilemmas in the Application of the

Safe Haven Principle

The "safe harbor rule" was created by the Digital

Millennium Copyright Act, enacted in the United

States, whose traditional "notice-and-takedown"

mechanism relies on manual review. However, short

video platforms actively intervene in the

dissemination of content through algorithms (e.g.,

dynamically adjusting search weights, implementing

personalized push), gradually building a new

governance framework of "algorithmic notification-

algorithmic deletion" and transferring the obligation

of censorship to algorithms. This shift has led to the

traditional "notice-and-takedown" mechanism falling

into an inefficient cycle - copyright owners are not

willing to defend their rights due to the difficulty of

proof and low compensation, while platforms have

formed a path of dependence and passively wait for

notices of infringement from rights holders (Li, 2025

& Xu & Wei, 2025). However, faced with a massive

amount of short video content, algorithms are

difficult to correctly and accurately identify fair use,

resulting in fair use content being mistakenly deleted,

and the anti-notification program, which should be a

relief mechanism, is difficult to play a substantive

role due to technical barriers and procedural idleness,

which objectively results in an imbalance between

2025).

3.2 The Contradiction Between

Platform Responsibility and

Technological Neutrality

The core competitiveness of the platform as an

information intermediary lies in the efficiency of its

algorithm, which can accurately recommend

commodities and contents through users' preferences

and browsing data to enhance the commercial value of

the platform. However, when recommending

contents, the platform may be driven by commercial

interests and prioritize the pushing of contents of

certain copyright holders while ignoring the rights and

interests of other copyright holders, to achieve

differentiated management. Such selective copyright

protection can lead to unfair market competition and

jeopardize the interests of copyright holders and users.

When faced with allegations of copyright

infringement, platforms often claim technological

neutrality and evade responsibility on the grounds that

the algorithms run automatically and that they cannot

control the results of their own recommendations

(Zhou, 2023). This has triggered a reflection on

technology neutrality, i.e., whether the technology is

neutral and whether the platform should take

responsibility for the results of algorithmic

recommendation. Algorithmic recommendation is not

a completely neutral technology, and will be driven by

the commercial interests of the platform, the platform

in order to maximize traffic and maximize commercial

interests, will be actively intervening in the

distribution of content through parameter adjustments.

For example, the platform may increase the

recommendation weight of certain popular content or

paid content, and reduce the probability of

recommendation of other content. This intervention

constitutes the basis for the determination of "should

know" infringement in Article 1195 of the Civil Code,

i.e., the platform should know that the content it

recommends may have the risk of infringement, but

still recommend it, and thus should bear the

responsibility for infringement. This shows that the

platform in the algorithmic recommendation is not a

passive technology provider, but a subject with

subjective intent and control ability, should bear the

duty of care. Based on the non-neutrality of

algorithmic recommendation, the obligation of the

platform should also be expanded. From the point of

view of foreseeable possibility, the platform, as the

developer and manager of the algorithm, should

foresee the risk of infringement and other negative

impacts that the algorithmic recommendation may

bring. From the point of view of control ability,

platforms can adjust and optimize the algorithms

through technical means, to control the

Study on Copyright Infringement Liability of Short Video Platforms from the Perspective of Intelligent Algorithm Recommendation

199

recommendation results. Therefore, platforms need to

be responsible for the recommendation results,

especially for popular content, and need to assume a

higher audit obligation. For example, platforms need

to adopt technical filtering (e.g. video fingerprinting)

and manual auditing for popular dramas and well-

known works to form a "selective" protection

mechanism.

3.3 Differences in Judicial Practice on

the Recognition of the Duty of Care

of Platforms

3.3.1 Typological Analysis of Typical Cases

The differences in the determination of the platform's

duty of care under the algorithmic recommendation

scenario are centrally manifested in the differences in

the weighing scale between the principle of

"technological neutrality" and the standard of "should

know" in judicial decisions. By sorting out the two

types of typical cases, it can be summarized into

"strict standard" and "loose standard" adjudication

path.

First, strict standards - take the case of "iQIYI v.

ByteDance" as an example (Beijing Haidian District

People's Court, 2018). In the case of iQIYI v.

ByteDance, the platform pushed a large number of

clips infringing on the copyright of Yanxi Raiders to

users through an algorithm. The court broke through

the traditional mechanism of "notification-deletion"

and held that the platform's failure to take the

necessary measures despite knowing the existence of

a large number of infringing behaviors by users was

an act of helping infringement. As Table 1 shows,it

also argued that the platform was "knowingly" at fault

from three aspects: technical interference,

commercial profitability, and technical feasibility.

The court's final decision rejected the platform's

"technology neutrality" defense and imposed a higher

duty of care on the platform. The case demonstrated

a strict standard of adjudication, and the judgment's

determination of the scope of the platform's "due

diligence" was also a source of controversy.

The second is a lenient standard - take the case of

"The General History of China in the Museum"

Property Court, 2024). In that case, the platform

recommended infringing documentaries uploaded by

users through a collaborative filtering algorithm, but

did not set specific recommendation rules. As Table

1 shows, the court strictly applied the standard of

"knowledge or substantial assistance", and held that

the platform's algorithm was generated automatically

based on the user's behavioral data, not as a result of

the platform's active intervention, with respect to the

infringing videos other than those repeatedly

uploaded

by the user "Duo Moumou". Some of the

Table 1: Comparison table of adjudication standards between the case of iQIYI v. ByteDance and the copyright dispute case

of "The General History of China in the Museum.

Dimension

Whether the platform

actively intervenes in the

technolog

Platform commercial

profitability

Feasibility

Scope of the Platform's

"due diligence"

iQIYI v.

Byte Jump

The platform actively

intervened with

technology that utilized

algorithms to actively

recommend infringing

content.

The platform profited

from actively

intervening by

recommending and

distributing infringing

content.

The platform had more

sophisticated copyright

filtering technology, but

didn't take the necessary

measures to block the

distribution of infringing

content.

The platform had

sufficient conditions,

capacity and reasonable

grounds to know that the

user had committed the

infringement in question.

dispute over

The General

History of

China in

Museums

The platform didn't

actively intervene in the

technology and only used

collaborative filtering

algorithms to recommend

and didn't set special

recommendation rules.

The platform's

algorithm was only

given to the automatic

generation of the user's

behavioral data, not the

result of the platform's

active intervention, and

the platform didn't have

the intention to profit

from infringing content.

The platform had more

mature copyright

filtering technology, but

didn't take action to stop

repeat infringement that

matched its distribution

capacity, but removed

other infringing videos in

a timely manner.

The platform's

algorithmic

recommendation was a

personalized

recommendation tailored

to the user's behavioral

data, which was different

from the judicial active

recommendation, and the

platform did not know

that the personalized

recommendation was an

infrin

video.

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

200

judgments in this case reflect the neutral and tolerant

stance of the judiciary towards algorithmic

recommendation technology, and reject the view that

platforms adopting algorithms should bear a higher

degree of responsibility.

3.3.2 Focus of Divergence in Court Decisions

Algorithmic recommendation technology has had an

impact on the traditional tort liability system. It is

mainly reflected in the differences in the

interpretation of the elements of "should know" and

"necessary measures" in Article 1197 of the Civil

Code. In judicial practice, there has been a large

discrepancy in the court's determination of the two

main elements. The determination of the platform's

duty of care and liability has also generated

considerable disagreement. The standard of "should

know" is based on the traditional "red flag standard",

that is, the fact of infringement is obvious and can be

easily recognized, but the emergence of algorithmic

recommendation technology makes it difficult to

apply this standard (Chen, 2023). Is there a

presumption that a platform "should have known"

because it uses algorithmic recommendation

techniques? If the algorithm has difficulty in

recognizing infringing videos when they are first

uploaded, does the platform have a "due diligence"

obligation? These questions require further study.

4 PATHS TO RECONFIGURE

THE BOUNDARIES OF

RESPONSIBILITY

4.1 Amendment of the Principle of

"Technology Neutrality"

Through the introduction of the "substantially

contributes to infringement" standard, if the

algorithm design significantly increases the risk of

infringement (e.g., collaborative filtering to

recommend popular infringing content), the defense

of neutrality will be negated. The "initiative" and

"purpose" of the algorithm design is the key to

determining liability. If the platform deliberately

guides users to infringing content through

collaborative filtering and heat weighting, its

behavior is beyond the scope of technological

neutrality. In the ByteDance case, for example, the

court found that the platform actively recommended

infringing short videos through algorithms, which

significantly improved the efficiency of content

dissemination and constituted a substantial promotion

of infringement and thus could not invoke the

principle of "technological neutrality" to exempt

itself from liability. In addition, the

"Recommendation System Transparency" clause in

the EU's Digital Services Act (DSA) requires

platforms to disclose the logic of their algorithmic

recommendations to assist courts in assessing

whether the algorithms tend to infringe. For example,

if the platform's algorithm gives priority to

unauthorized film and television clips, even if the

content is not directly uploaded, the algorithm may be

liable for indirect infringement due to the "induced"

design of the algorithm.

Judging the platform's proof of subjective fault in

the judiciary can be aided by algorithmic

transparency requirements for judicial

determinations. In terms of the algorithm filing

system, platforms are required to file core

recommended algorithm parameters and update

records with the regulatory authorities to ensure

traceability after the fact. For example, China's

"Regulations on the Administration of Algorithm

Recommendation for Internet Information Services"

have made clear the filing requirements, which can be

extended to copyright infringement scenarios. In

terms of the obligation to provide an explanatory

report, in infringement litigation, platforms are

required to submit an explanatory report on the

algorithm's decision-making logic, proving that they

have taken reasonable measures to avoid the

proliferation of infringing content. If the report

contains logical contradictions or avoids key issues,

the platform is presumed to be at fault.

In terms of third-party technical audits, a third-

party independent technical organization can be

introduced to conduct compliance audits of

algorithms, focusing on evaluating the effectiveness

of their copyright filtering measures, such as

comparison coverage, false positive rate and other

indicators.

4.2 Setting a Dynamic Standard of

Duty of Care

4.2.1 Tiered Model of Obligations

According to Article 1195 of the Civil Code, based

on the provisions of network infringement liability,

combined with Article 42 of the E-Commerce Law,

"Notice-Delete" rule, the platform liability is

positively correlated with the algorithmic control and

Article 24 of the Cybersecurity Law, "Necessary

Limits of Technological Measures", should be

Study on Copyright Infringement Liability of Short Video Platforms from the Perspective of Intelligent Algorithm Recommendation

201

adopted. The platform's liability is positively related

to the control of the algorithm and the "necessary

limit of technical measures" in Article 24 of the

Cybersecurity Law. Therefore, the platform should

make a comprehensive judgment on the control of the

algorithm (e.g. whether to actively set up the

recommendation rules), the heat of the infringing

content, and the record of repeated infringement. At

the first level, the platform designs the algorithmic

rules completely independently, but the platform

needs to assume the obligation of prior filtering (e.g.,

deploying copyright fingerprinting system) and real-

time monitoring of hot content. At the second level,

third-party algorithm services are used, but the

compliance of the third-party algorithms must be

audited, and reports on the handling of infringing

content must be submitted regularly. At the third

level, the platform only provides basic

recommendation functions and fulfills the obligation

of "notification and deletion" but needs to take

measures to restrict or block repeated infringing

users.

4.2.2 Technical Feasibility Considerations

Based on the feasibility of technology and the

effectiveness of filtering, platforms should adopt

filtering measures (such as keyword shielding and

algorithmic capabilities. In small and medium-sized

platforms, basic measures such as keyword shielding

and MD5 hash value comparison can be implemented

for filtering, while in large-sized platforms, AI image

recognition and audio fingerprinting technologies

(e.g., YouTube Content ID) can be deployed to

realize accurate interception of infringing content, or

to explore the automatic authorization of blockchain

deposits and smart contracts, so as to build an

ecosystem of full-chain copyright protection of

"creation-dissemination-vindication" or explore

blockchain certificate and smart contract automatic

authorization, to build a "creation-dissemination-

rights defense" full-chain copyright protection

ecology.

4.3 Construction of a

Multi-Dimensional Co-Governance

Mechanism

The reconfiguration of the liability boundary should

consider the balance between copyright protection

and technological development. As for the platform,

as the beneficiary of the algorithm, the platform

should bear the main responsibility for data security

according to Article 9 of the Data Security Law, and

should also avoid the "one-size-fits-all" type of

censure, and be allowed to gradually optimize the

algorithm within the scope of technical feasibility; for

the right holders, the exceptions to the "Safe Harbor

Principle" should be perfected. For rights holders, it

is necessary to improve the exceptions to the "safe

haven principle" and clarify the standard of proof for

substantive infringement; for users, it is also

necessary to protect the reasonable use of space, and

to avoid excessive filtering to inhibit secondary

creativity and cultural exchanges. Through the

establishment of a copyright pre-authorization

database and an efficient notification and deletion

mechanism, the cooperation between platforms and

right holders can be realized, while promoting the

intervention of algorithmic ethical review and third-

party technical assessment. In the future, market-

based mechanisms such as "algorithmic liability

insurance" can be explored to diversify the

compliance risk of platforms, while promoting the

formation of technical autonomy standards in the

industry.

5 CONCLUSION

At a time when intelligent algorithmic

recommendation is booming, the issue of copyright

infringement liability of short video platforms has

become more and more prominent. From amending

the principle of "technology neutrality", to

establishing a dynamic duty of care standard, to

improving the infringement relief mechanism, this

series of initiatives aims to balance the relationship

between technological innovation and copyright

protection. However, as technology continues to

evolve, new forms of infringement and complex

issues will continue to emerge. In the context of

globalization, the dissemination of short videos

knows no boundaries, so international cooperation in

short video copyright protection will also become

increasingly close. Countries need to strengthen

exchanges and collaboration, jointly formulate

internationally accepted rules and standards for short

video copyright protection, combat cross-border

infringement, and create a favorable international

environment for the healthy development of the short

video industry. In the future, it is necessary to pay

continuous attention to the development of the

industry and continuously improve the relevant legal

system and governance measures, to prompt short

video platforms to continue to innovate on a legal and

compliant track, and to realize a win-win situation

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

202

between technological progress and copyright

protection.

AUTHORS CONTRIBUTION

All the authors contributed equally and their names

were listed in alphabetical order.

REFERENCES

B. Chen, 2023. Multi-dimensional thinking of platform

algorithmic regulation: from case-by-case infringement

adjudication to algorithmic comprehensive governance.

L. A. J. 3, 91-94.

Beijing Haidian District People's Court. (2021, December

31). Civil Judgment [(2018) Jing 0108 Min Chu No.

49421].

https://www.sohu.com/a/518790526_121123754

Beijing Intellectual Property Court. (2024, December 19).

Civil Judgment [(2024) Jing 73 Min Zhong No.180].

C. Y. Le, Z. X. Wang, J. P. Zhang, et al. 2025. Research on

user algorithm response behavior under intelligent

recommendation of short-video platforms. Libr. Dev. 4,

1-28.

Short Video Platform in Intelligent Communication and

Its Governance Path--An Appraisal of the First Case of

Algorithmic Recommendation. S. J. R. 9, 86-96.

H. Zhou, 2021. Governance of Imbalance of Interests in

Single Market Copyright Directive. Intellect. Prop. 5,

41-55.

J. M. Sin, 2024. Controversy and Optimization of Judicial

Determination of Copyright Duty of Care for Short

Video Platforms under Algorithmic Recommendation.

J. S. U. (P. S. S.). 26, S1, 106-111.

Protection Behavior of Short Video Platform

Algorithms. S. S. I. G. 2, 38-52, 286.

and Data Mining in Big Data Environment - A

Perspective on the EU's Proposal for the DSM

S. H. Zhou, 2023. "The first case of algorithmic

recommendation" in the short video platform's duty of

care determination study. S. S. I. G. 4, 64-72.

X. Y. Li, 2025. Systematic Regulation of Algorithmic

Recommendation Infringement on Short Video

Platforms. H. L. S. 6, 147-163.

Y. Wan, 2021. China's Choice of Mandatory Filtering

Mechanism in Copyright Law. S. I. L. B. 6, 184-196.

Study on Copyright Infringement Liability of Short Video Platforms from the Perspective of Intelligent Algorithm Recommendation

203