Criminal Regulation of AI Crimes: The Application Dilemma and

Path Optimization of the Crime of Assisting Information Network

Criminal Activities

Jiaxin Shi

School of Foreign Languages and Literature, Wuhan University, Wuhan, China

Keywords: AI Crimes, Crime of Assisting Information Network Criminal Activities, Criminal Regulation.

Abstract: With the rapid development of artificial intelligence technology, its wide application in various fields of

society not only promotes efficiency innovation but also provides technical tools for new types of crimes.

Crime forms such as AI-driven automated fraud, deepfake, and data theft, with their concealment, high ef-

ficiency, and cross-domain nature, pose severe challenges to the traditional criminal law system. This pa-per

focuses on Article 287(2) of the Criminal Law, "Crime of Assisting Information Network Criminal

Activities", and explores its applicability in AI crimes. In judicial practice, this crime faces many dilem-mas.

For example, the principle of technological neutrality makes the definition of criminal acts vague, the

determination of subjective knowledge is difficult in the face of complex AI application scenarios, and the

division of responsibilities is unclear when multiple parties are involved. This paper deeply ana-lyzes these

core controversial issues and proposes suggestions for improving legal regulation and indus-try supervision,

aiming to provide theoretical support for constructing a criminal governance framework adapted to the AI

era.

1 INTRODUCTION

Driven by the digital wave, artificial intelligence (AI)

technology has penetrated into all fields of social life,

from medical diagnosis to financial transactions, from

public opinion analysis to automated production. Its

high efficiency and innovation have injected strong

impetus into the development of human society.

However, the "double-edged sword" effect of

technology has also become apparent in this process.

The algorithmic autonomy, data dependence, and

technological generalization of AI are being used by

criminals as new tools for committing crimes.

Deepfake technology can generate lifelike audio and

video, which can be used for slander, fraud, or

interference in judicial activities. Automated

algorithms push fraudulent information through

precise profiling, causing the scale of victims to grow

exponentially. Malicious programs, with the rapid

iteration ability of AI, pose a systematic threat to the

network security defense line. Such crimes not only

violate personal privacy and property security but

also impact the social trust foundation, disrupt the

economic order, and even threaten national security.

Facing this challenge, the traditional criminal law

system is in a dilemma in terms of the identification

of the subject of behavior, the proof of subjective

intent, and the evaluation of technological neutrality,

and urgently needs to explore regulatory paths that

adapt to the characteristics of AI crimes.

Currently, domestic and foreign academic circles

have initially explored the criminal regulation of AI

crimes. Domestic research focuses on the applicable

boundaries of specific crimes. For example, regarding

the controversy over Article 287(2) of the Criminal

Law, "Crime of Assisting Information Network

Criminal Activities," scholars have analyzed the

"knowledge" standard, the scope of assisting acts, and

the severity of circumstances in its constitutive

elements from the perspectives of norm theory and

dogmatics（Zeng & Jin, 2024). Some viewpoints

advocate narrowing the application scope of this

crime to avoid over-generalization (Wu & Zhang,

2023). There are also studies that propose to solve the

identification problems through the coordination of

technical means and legal interpretation, combined

with judicial practices such as the "Breaking the

Card" operation (Wei & Sha, 2023).

In contrast, international research pays more

attention to the global and technical characteristics of

AI-driven crimes. Scholars such as Chanderprajapu

point out that the existing international legal

Shi, J.

Criminal Regulation of AI Crimes: The Application Dilemma and Path Optimization of the Crime of Assisting Information Network Criminal Activities.

DOI: 10.5220/0014293300004859

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 27-32

ISBN: 978-989-758-785-6

framework has problems of ambiguous jurisdiction

and broken evidence chains in dealing with AI-

driven-cyber-crimes. The United Nations and Interpol

call for strengthening governance through

transnational cooperation and the unification of

technical standards. However, existing research

mostly focuses on a single legal provision or technical

scenario, lacks a systematic response to core

controversial issues such as the liability

determination of "technology providers" in AI crimes

and the nature definition of algorithmic

recommendation behavior, and has not formed a

regulatory system that takes into account both

technological development and risk prevention and

control.

This paper takes Article 287(2) of the Criminal

Law, "Crime of Assisting Information Network

Criminal Activities," as the starting point to deeply

explore its application dilemma and optimization path

in AI crimes. First, the article sorts out the typical

types of AI crimes (such as automated fraud, data

theft, and deepfake) and their multi-dimensional

harms to social governance, revealing the

shortcomings of the traditional criminal law in

dealing with such crimes. Then, from the perspective

of legal hermeneutics, it analyzes the constitutive

elements of this crime, focusing on the determination

standard of "subjective knowledge," the boundary

division of technologically neutral behavior, and the

impact of AI's autonomous decision-making on

liability attribution. Finally, combined with

technological characteristics and judicial practice, it

proposes systematic suggestions such as clarifying

the boundaries of responsible subjects, refining the

rules for determining "knowledge," strengthening

industry supervision, and formulating special judicial

interpretations, aiming to provide theoretical support

for constructing a criminal governance framework

that is technically adaptable and has clear rights and

responsibilities.

The full text is divided into five parts: The first

part (Introduction) clarifies the necessity and

innovation of the research through background

analysis and literature review; the second part

systematically summarizes the main types, technical

characteristics, and harm evolution trends of AI

crimes; the third part, based on legal texts and judicial

interpretations, analyzes the constitutive elements

and judicial identification difficulties of the "Crime of

Assisting Information Network Criminal Activities";

the fourth part focuses on the application

controversies of this crime in AI crimes, including the

ambiguous liability of technology providers, the

doubtful nature of algorithmic recommendation, and

the impact of AI autonomy on subjective intent; the

fifth part proposes improvement paths from the

legislative, judicial, and industry supervision levels,

emphasizing the realization of a balance between risk

prevention and control and innovation protection

through "technology - law" coordinated governance.

This paper attempts to bridge the fragmented defects

of existing research and provide reference for the

scientific and refined criminal legislation in the AI

era.

2 TYPES AND DEVELOPMENT

TRENDS OF AI CRIMES

2.1 Typical Application Types

In current criminal activities, the application of AI is

becoming more diverse and concealed. Typical examples of

AI crimes include automated fraud, data theft crimes, and

deepfake technology. Among them, automated fraud uses

AI algorithms to accurately profile a large number of

potential victims. According to factors such as different

people's psychological characteristics and consumption

habits, personalized fraud scripts are customized and

widely spread through channels such as text messages,

phone calls, and online advertisements. In data theft crimes,

AI technology can quickly analyze the vulnerabilities of

network systems and automatically capture massive

amounts of personal information and business - confidential

data, providing data support for subsequent illegal

transactions or criminal activities. Deepfake technology

creates false content that is enough to be confused with the

real by deeply learning a large number of image, audio, and

video materials, which is used for illegal and criminal acts

such as slandering others, creating social panic, or

interfering with judicial procedures.

2.2 Impact on Social Harmfulness

The social harmfulness of AI crimes cannot be

underestimated, which is mainly reflected in the

economic level, social order level, and national

security level. From an economic perspective,

automated fraud and data theft directly lead to a large

amount of property losses. Enterprises may face

problems such as damage to business reputation and

economic compensation due to data leakage, which in

turn affects the stable development of the industry. In

terms of social order, the false information of

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

deepfake is extremely likely to cause public panic,

disrupt the normal social public opinion environment,

mislead public perception, and may even trigger mass

incidents. In addition, AI crimes also pose a serious

threat to national information security. The leakage of

sensitive data may be used for espionage activities,

cyber - attacks, etc., damaging national interests and

international image.

3 LEGAL ANALYSIS OF THE

CRIME OF ASSISTING

INFORMATION NETWORK

CRIMINAL ACTIVITIES

3.1 Analysis of Constitutive Elements

According to Article 287(2) of the Criminal Law and

relevant judicial interpretations, from the perspective

of the four constitutive elements, the constitutive

elements of the crime of assisting information

network criminal activities include the subject,

subjective aspect, object, and objective aspect. The

subject is a general subject, including natural persons

and units. The subjective aspect requires that the actor

knows that others are using the information network

to commit crimes and still provides technical support,

advertising promotion, payment settlement, and other

assisting acts. The object is a complex object, which

not only violates the management order of

information network security but also infringes the

legitimate rights and interests of citizens, legal

persons, and other organizations. The objective aspect

is manifested as providing assistance for information

network crimes and reaching the level of serious

circumstances.

3.2 Discussion on Key Identification

Standards

The determination of the "knowledge" standard is the

key to the application of this crime. As the legal

proverb goes, a single piece of evidence cannot

establish a fact. In practice, it cannot be judged only

based on the actor's own confession. Factors such as

the actor's cognitive ability, the channels of accessing

information, and the abnormality of the behavior need

to be comprehensively considered. Regarding the

scope of "assisting acts," with the development of

network technology, it not only includes traditional

technical support, payment settlement, and other acts

but also covers new network services such as cloud

computing services and algorithmic recommendation

services. The definition of "serious circumstances"

needs to be judged by combining multiple factors

such as the number of assisting acts, the duration, and

the resulting harm consequences.

4 APPLICATION DILEMMA OF

THE CRIME OF ASSISTING

INFORMATION NETWORK

CRIMINAL ACTIVITIES IN AI

CRIMES

4.1 Liability Determination of AI Tool

Providers

In AI crimes, there is a controversy over whether AI

tool providers constitute "assistants." Regarding

whether AI tool providers constitute "assistants,"

there is a controversy in the academic community.

Some scholars believe that the behavior of AI tool

providers is essentially a neutral technical behavior

and should not be directly regarded as an assisting act

(Yuan & Xue, 2024). For example, the provider of a

generative AI tool may only provide a technical

platform and cannot foresee the specific criminal acts

for which its tool is used. Some AI tools are universal,

and it is difficult for the provider to know the specific

uses of the users (Jiang & Liu, 2024).

If all tool providers are identified as accessory

offenders, it may limit the normal development of AI

technology; but if not regulated, a large number of

tools may be used for criminal activities.

4.2 Industry Supervision Failure

In the author's opinion, the problem of industry

supervision failure can be mainly manifested in the

following three aspects.

First, inequality between supervision

responsibilities and powers. Currently, private

enterprises assume a large number of obligations in

network space supervision but lack corresponding

power support. For example, laws such as the Anti-

telecommunications Fraud Law of the People's

Republic of China clearly define the responsible

subjects of new business fraud-related risk safety

assessment, risk prevention and control, internal

control mechanisms, and security responsibility

systems as telecommunications business operators,

Criminal Regulation of AI Crimes: The Application Dilemma and Path Optimization of the Crime of Assisting Information Network

Criminal Activities

banking financial institutions, Internet service

providers, etc. However, these enterprises do not have

the power to independently decide to restrict, close,

prohibit, or freeze the network activities of others.

Their supervision responsibility is more of an

obligation to the main body of network activities

rather than real power. This inequality between

responsibility and power makes it difficult for

enterprises to fulfill their supervision obligations and

also increases their risk of being held criminally liable

for insufficient supervision.

Second, mismatch between technological

development and supervision capabilities. The

government lags behind in the research and

development and application of network governance

technologies and is difficult to effectively respond to

the rapidly developing information technology and AI

technology. Although the government is nominally

the responsible subject of network governance, it

lacks the actual ability to lead the development of

network governance technologies. This mismatch

between technological development and supervision

capabilities leads to the over-reliance on the role of

criminal law in governing network crimes and

increases the risk of criminal law being abused.

Third, incompleteness of industry supervision

standards. Currently, there are still deficiencies in the

industry supervision standards for AI technology. For

example, although the Interim Measures for the

Administration of Generative AI Services put forward

compliance supervision requirements for the

generative AI-related industries, these requirements

mainly focus on the technical level, such as adding a

differentiation mark to the generated images or videos

and standardizing the data collection behavior of

personal information. However, there is still a lack of

clear legal guidance on how to determine the

"knowledge" of platform or technical service

providers when they know that the other party is using

their technology for criminal activities and how to

assume criminal liability.

4.3 Impact of AI System's Autonomous

Decision-Making on "Subjective

Knowledge"

The autonomous decision-making ability of an AI

system may affect the determination of its

"knowledge" state. If an AI system can independently

judge and choose criminal acts, it may be determined

to have "knowledge." However, if the decision-

making of an AI system is completely based on preset

algorithms and data and does not show subjective

intent, it is difficult to determine that it has

"knowledge." The autonomous decision - making

process of an AI system is often highly complex and

unexplainable, especially for AI systems based on

deep learning and large models, whose internal

algorithms and decision-making mechanisms are

regarded as "black boxes." This unexplainability

poses a huge challenge in judging whether the actor

"knows" that his behavior may lead to criminal

results. For example, in AI-assisted financial fraud or

cyber - crimes, even if the actor uses an AI tool, it is

difficult to prove his subjective knowledge of the

criminal results through traditional method

5 SUGGESTIONS FOR

IMPROVING THE CRIMINAL

REGULATION OF AI CRIMES

5.1 Clarify the Liability Boundaries of

AI Tool Providers

The complexity and technical dependence of AI

crimes require clear definition of the liability

boundaries of AI tool providers and the realization of

operability in liability determination through

technological transparency. On the one hand, it is

necessary to clearly divide the liability levels of

different subjects. The development, deployment, and

application of AI technology involve multiple parties.

Developers, manufacturers, and users may jointly

cause criminal consequences due to technical defects

or abuse. For example, if a developer deliberately

reserves data-stealing loopholes in the design of an

algorithm or knows that the technology may be used

for illegal purposes but does not set up a risk -

blocking mechanism, he should bear the main

criminal liability; if the manufacturer fails to fulfill

the quality review obligation for AI products with

potential safety hazards, it should bear joint liability;

and when the user maliciously uses the technology to

commit a crime, he should bear independent liability.

By distinguishing the degree of fault and the nature of

the behavior of each subject in the technology chain,

it can avoid the suppression of technological

innovation by "one-size-fits - all" imputation and

accurately combat the source of crimes. On the other

hand, it is necessary to strengthen the requirements

for algorithm transparency and interpretability.

Currently, due to the "black-box" characteristics of

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

some AI systems, their decision - making logic is

difficult to trace, and developers often shirk their

responsibilities on the grounds of technological

neutrality. In this regard, legislation should be used to

force high-risk AI tools (such as deepfake programs,

automated recommendation systems) to disclose the

core algorithm framework and embed an

interpretability module. For example, attach a

"technical path description" to the output of

generative AI, recording the data source and decision-

making basis; introduce a third-party audit

mechanism in the algorithm training stage to ensure

that the development process complies with ethical

norms. Only by breaking through the technical

barriers can objective bases be provided for the

determination of "subjective knowledge" and

"assisting acts" in judicial practice and the balance

between the principle of technological neutrality and

the principle of liability adaptation be achieved.

5.2 Strengthen Industry Supervision

The rapid iteration and wide application of AI

technology require that industry supervision must

shift from passive response to active prevention and

control. To achieve this goal, it is necessary to build a

trinity supervision framework of "legal norms-

enterprise self-discipline-technical support" and form

a risk prevention and control system covering the

entire life cycle of AI through multi-level system

design, criminal compliance incentives, and the

coordination of technology and law.

First, establish a multi-level three-dimensional

supervision system to achieve comprehensive

coverage from legislation to practice. At the

legislative level, special laws such as the AI Law

should be accelerated. The bottom-line requirements

for technology research and development, data use,

and product deployment should be clarified. For

example, it should be stipulated that the design of AI

tools must embed an ethical review mechanism, and

the development of algorithm models with obvious

criminal orientation should be prohibited. At the same

time, industry access standards should be refined

through administrative regulations, and enterprises

should be required to complete safety assessment and

filing before entering the market to ensure the legality

of technology application. At the administrative

supervision level, it is necessary to strengthen the

cross-departmental cooperation mechanism. For

example, the Internet Information Office, the public

security organ, and the science and technology

management institution jointly establish an "AI

Safety Supervision Committee" to regularly conduct

special inspections on high - risk areas (such as

deepfake, automated recommendation systems) and

impose dynamic penalties on illegal enterprises.

Technical supervision needs to rely on third - party

testing institutions to conduct transparent reviews of

the operation logic and output results of AI systems

through technical means such as algorithm auditing

and data traceability to avoid supervision blind spots

caused by "black-box operations."

Second, promote the corporate criminal

compliance plan and internalize risk prevention and

control into the conscious actions of industry

development. The practical experience of the EU's AI

Liability Directive can be borrowed to require AI

enterprises to establish a compliance management

system including risk identification, internal control,

and emergency response. For example, when an

enterprise develops a face recognition system, it

needs to pre-evaluate the risk that it may be used for

illegal monitoring or identity theft and embed a

"usage scenario restriction" function in the algorithm;

in the data collection link, user explicit consent and

anonymization processing should be adopted to avoid

privacy violations. For enterprises that actively fulfill

compliance obligations, policy incentives such as tax

reduction and priority in market access can be given;

conversely, for enterprises that allow the abuse of

technology, administrative penalty intensity should

be increased, and even the criminal liability of

relevant responsible persons should be investigated.

In addition, industry associations should take the lead

in formulating the AI Ethical Guidelines to guide

enterprises to integrate the concept of "technology for

good" into the entire product process. For example,

generative AI tools are required to mark "deepfake

risk warnings" to reduce the possibility of technology

abuse from the source.

Third, strengthen the in-depth integration of

technology and law to improve the accuracy and

efficiency of judicial governance. On the one hand, it

is necessary to improve the cognitive level of judicial

personnel on AI technology through professional

training and interdisciplinary cooperation. For

example, when hearing cases involving algorithmic

recommendation, judges should master basic

machine learning principles and be able to distinguish

between "technically neutral push" and "malicious

inducement behavior"; when determining "subjective

knowledge," technical experts can be used to analyze

system logs and algorithm parameters to judge

Criminal Regulation of AI Crimes: The Application Dilemma and Path Optimization of the Crime of Assisting Information Network

Criminal Activities

whether the developer has criminal intent. On the

other hand, the auxiliary application of AI technology

in judicial practice should be explored. For example,

natural language processing technology can be used

to analyze a large number of judgment documents to

extract the judgment rules of the "Crime of Assisting

Information Network Criminal.

6 CONCLUSION

The rapid advancement of artificial intelligence (AI)

technology has brought unprecedented opportunities

and challenges to society. While AI continues to drive

innovation and efficiency, its misuse in criminal

activities has posed significant threats to personal

privacy, economic stability, social order, and national

security. This paper has explored the application

dilemmas and optimization paths of Article 287(2) of

the Criminal Law, "Crime of Assisting Information

Network Criminal Activities," in the context of AI

crimes.

Through an analysis of typical AI crime types,

such as automated fraud, data theft, and deepfake

technology, the paper highlights the multidimensional

harms of AI crimes and the shortcomings of

traditional criminal law in addressing these

challenges. It further examines the key issues in the

application of Article 287(2), including the

determination of subjective knowledge, the boundary

of technologically neutral behavior, and the impact of

AI autonomy on liability attribution.

The paper argues that the complexity and

technical nature of AI crimes necessitate a balanced

approach to criminal regulation, emphasizing both

risk prevention and innovation protection. To achieve

this, the paper proposes several recommendations:

clarifying the liability boundaries of AI tool

providers, strengthening industry supervision through

a "law-technology" coordinated governance

framework, and enhancing the transparency and

interpretability of algorithms. These measures aim to

address the challenges posed by AI crimes while

fostering the responsible development of AI

technology.

In conclusion, this paper provides a

comprehensive framework for understanding and

regulating AI crimes, offering theoretical support for

the construction of a criminal governance framework

adapted to the AI era. By addressing the core

controversies and proposing practical solutions, the

paper contributes to the ongoing discourse on AI

governance and legal regulation.

REFERENCES

Jiang Luoyi, Liu Shude. 2024. Examination and Regulation

of Judicial Reasoning on “Knowledge” in the Crime of

Assisting Information Network Criminal Activities.

Research on the Modernization of Rule of Law, 5:124-

135.

Wei Hanxi, Sha Guijun. 2023. Practical Dilemmas and

Countermeasures in the Judicial Application of the

Crime of Assisting Information Network Criminal

Activities. Journal of the Criminal Investigation Police

University of China 6:57-65.

Wu Dianchao, Zhang Yu. 2023. Research on Several Issues

in the Application of the Crime of Assisting Information

Network Criminal Activities. Journal of Henan

Institute of Education (Philosophy and Social Sciences

Edition) 3:54-57.

Yuan Bin, Xue Liming. 2024. Research on the Criminal

Law Regulation of Generative Artificial Intelligence.

Hebei Law Science 2:140-159.

Zeng Lei, Jin Yuanyuan. 2024. The Normative Attributes

and Interpretive Rectification of the Crime of Assisting

Information Network Criminal Activities. Journal of

Chongqing University of Posts and

Telecommunications (Social Science Edition) 3:52-61.

Zhang Mingkai. 2016. On the Crime of Helping

Information Network Crime. Political Science and Law

2:2-16.

Pi Yong. 2018. On the Legislation of New Types of

Cybercrime and Its Application. Social Sciences in

China 10:126-150.

Liu Yanhong. 2016. A Critique of the Criminalization of

Cybercrime Aiding Behaviors as Principal Offenses.

Law and Business Research 3:18-22.

Li Hong. 2017. On the Nature and Application of the Crime

of Assisting Information Network Criminal Activities.

Legal Application 21:33-39.

Jiang Su. 2020. The Interpretive Direction of the Crime of

Assisting Information Network Criminal Activities.

China Criminal Law Journal 5:76-93.

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