Artificial Intelligence in China’s Banking Sector: Promises, Perils, and Regulation

Amidst the explosive growth of Artificial Intelligence (AI) in China in recent years, Chinese financial market participants, particularly banks, have proactively sought to integrate AI into their operations. Marked by landmark policies, specifically the ‘Action Plan for Promoting High-Quality Development of Digital Finance’ spearheaded by the People’s Bank of China (PBOC), China’s approach to AI in finance is profoundly shaped by two fundamental dimensions. 

First, China’s rich data and advancing financial technology (FinTech) infrastructure establishes robust data foundations for AI development. Data, as the core production element of AI technology, directly determines the scope and efficacy of algorithmic models. Bolstered by its extensive population and a series of strategic national policies, China possesses significant data advantages for developing and training AI systems. In this regard, the ‘Financial Technology Development Plan (2022-2025)’ issued by the PBOC sets ambitious goals for the full release of data element value and high-quality digital transformation, aiming to achieve a ‘digital, intelligent, green, and fair’ financial service capability by 2025.

Parallel to this, China’s FinTech sector has demonstrated dynamic growth. By 2021, the core AI industry had reached a scale exceeding 180 billion RMB, catalyzing related industries valued at over 740 billion RMB.[i] In the first half of 2025, generative AI technologies in China witnessed substantial advancements, reflected in the rapid increase in product development and the diversification of application scenarios.

At the institutional level, persistent increases in IT investment by financial institutions underscore the strength of this technological transformation, rising from 359.77 billion RMB in 2023 to 394.96 billion RMB in 2024. These technological advancements have equipped financial institutions, particularly banks, with the capability to deploy large-scale AI applications more effectively. The burgeoning market for financial large models[ii] in China is indicative of this trend, with the market size expected to surge from 1.593 billion RMB in 2023 to 13.179 billion RMB by 2028.

Second, from a regulatory perspective, China, as the world’s second-largest economy, and characterized by a large population and regional disparities, has adopted an innovation-friendly strategy and an adaptive regulatory framework for FinTech as part of its drive towards inclusive and sustainable finance of the nation. While the banking sector plays an essential role in China’s financial system, its current regulatory strategy towards AI focuses on encouraging financial innovation alongside risk management, enabling the rapid deployment and continuous iteration of AI within the industry. This ‘develop first’ approach stands in sharp contrast to the European Union’s (EU) ‘regulation first’ model and the more laissez-faire innovation paradigm in the United States (US).

While China’s rapid deployment of AI in the financial markets has driven significant advancements, it has also introduced substantial risks that require careful regulatory responses. At the technical level, issues such as computational errors and the ‘black box’ opacity of algorithmic models present persistent challenges. At the market level, the widespread use of similar algorithms can increase risk concentration and pro-cyclicality, potentially resulting in model resonance risks. 

A central challenge for both regulators and market participants is effectively managing these risks while fostering technological innovation. Although a growing body of literature and numerous policy reports have examined AI in finance, a distinct research gap remains. There is a scarcity of academic analysis that systematically examines the latest laws and practices in China’s banking sector from a legal perspective. 

My forthcoming paper  titled ‘Artificial Intelligence in China’s Banking Sector: Promises, Perils, and Regulation’ seeks to fill the literature gap by addressing three key questions: (1) how is AI being integrated into the banking sector? (2) what are the major risks of AI banking? (3) what roles should regulators play in addressing risks associated with AI adoption in the banking sector? 

The article first discusses the factors driving the rapid adoption of AI technology by banks in China and clarifies the governance pathway that has emerged, shaped by the country’s distinctive market structure and regulatory approach. It demonstrates that China’s strong policy emphasis on inclusive finance, coupled with the accelerated adoption of digital and AI technologies, offers new avenues for overcoming the Financial Inclusion Trilemma of scale, risk, and profitability in inclusive finance .

The second part of the paper details the major players and specific application scenarios of AI within the banking industry. It is found that AI offers an alternative solution to address the ‘Financial Inclusion Trilemma’ by enhancing efficiency and expanding access. However, it also introduces complex risks ranging from data integrity issues and model opacity to systemic vulnerabilities. These are risks that traditional regulatory frameworks may be ill-equipped to manage. Consequently, the role of regulation is increasingly critical in safeguarding financial stability without stifling the innovation that fuels economic growth.

Navigating this landscape may necessitate a fundamental shift in regulatory capability and strategy. First, given the high uncertainty and new risks posed by AI, the regulators should adopt a risk-based and adaptive framework for AI in banking. Such an approach can begin with high-level principles or codes issued as soft law, translate these into practical methods and toolkits through industry pilot programmes, handbooks and guidelines, and ultimately evolve into hard law with legal effect (such as Rules or Regulations on AI Risk Management). 

Moreover, traditional static regulation should evolve toward dynamic and technology-enabled oversight. In this context, the deployment of RegTech by financial institutions and SupTech by regulators is a vital component in providing real-time visibility needed to monitor ‘black box’ algorithms. However, technology alone may be insufficient if not matched by human capital. Enhancing regulatory capacity arguably requires a concerted effort to attract multidisciplinary talent. At the same time, it remains important for financial institutions to elevate the sophistication of their internal governance, potentially by empowering roles such as CTOs and dedicated Chief AI Officers. Institutional reforms within regulatory bodies, similar to the Monetary Authority of Singapore’s (MAS) Fintech & Innovation Group, could serve as a model for fostering this dual capacity of supervision and innovation facilitation.

Furthermore, to balance financial stability with innovation in a rapidly evolving banking sector, the regulatory framework should be technology-neutral, proportionate to use-case risk, embedded within existing risk regimes, and refined through close consultation with financial institutions and evidence from practice. Beyond domestic regulatory reforms and efforts, international cooperation remains important for establishing consistent global standards. As illustrated by the strategic partnership between the MAS and the UK Financial Conduct Authority (FCA), such cross-border collaboration may serve as a vital mechanism to help harmonize expectations and facilitate the realization of AI’s transformative potential within a secure banking system. 

The author’s paper can be found here. 

Lin Lin is an Associate Professor at Faculty of Law of the National University of Singapore.  

Endnotes

[i]       Lu Y, Li B, Shi H, ‘The Development Path of China’s FinTech Industry: Models, Issues, and Regulatory Mechanisms’ 128 Comparison 234-263 (2023).

[ii]      A large model or large-scale model refers to one trained on substantial data with a complex computational architecture, capable of handling complex tasks and possessing generalization ability, typically with no fewer than 100 million parameters. In contrast, small models are lighter, task-specific, and easier to deploy. See China Academy of Information and Communications Technology (2024), p. 32.