The advent of cutting-edge artificial intelligence technologies has prompted many banks to actively integrate advanced AI solutions into their operationsA prominent example is DeepSeek, an AI model that excels in data processing, natural language understanding, and logical reasoningIts versatility makes it suitable for a variety of critical banking functions, including credit checking, contract management, and customer serviceHowever, with the rapid adoption of AI, banks are increasingly focused on the potential risks, including concerns surrounding data security, model inaccuracies, and regulatory compliance.

Recently, Liu Tong, the Deputy General Manager of the Product Center at China Financial Certification Center Co., Ltd. (CFCA), shared insights into these developments during an exclusive interviewHe emphasized that while DeepSeek offers a risk-controlled intelligent solution for the banking sector, its successful implementation requires a careful balance between technological innovation and security complianceBanks must leverage the advantages of AI while simultaneously fortifying their defenses against potential threats to financial security.

The shift from mere efficiency improvements to an expansive ecological transformation within the banking industry is evident as multiple banks have begun deploying the DeepSeek model in localized settingsThis deployment encompasses various applications such as smart contract management, intelligent risk control, asset custody and reconciliation, customer service assistance, and research tasks.

Nonetheless, several significant challenges persist in integrating AI into banking: the imbalance between training costs and energy efficiency, insufficient generalization capabilities for rare scenarios, and the engineering hurdles of deploying AI models at the edgeLiu Tong posits that DeepSeek is effectively tackling these issues through optimized cost structures and compatibility with domestically produced technology stacks.

To illustrate DeepSeek's impact, Liu notes its advantages in enhancing the efficiency, precision, and risk management capabilities of banks, ultimately driving the modernization of essential infrastructure

Advertisements

In critical areas such as lending, contract evaluation, customer support, and reconciliation, DeepSeek offers clear opportunities for immediate integration.

In the realm of credit assessment, traditional approaches often rely on manual processing of unstructured data, which is time-consuming and error-proneIn contrast, DeepSeek's superior understanding of the Chinese language and its adept logical reasoning capabilities can automate the parsing of credit materials, thus significantly expediting the review processAdditionally, in contract management, DeepSeek utilizes natural language processing (NLP) technologies to accurately identify contract clauses and conduct compliance checks against legal databasesThis automated functionality not only reduces human error but also streamlines the compliance verification process.

Furthermore, DeepSeek enhances customer service by offering personalized assistance based on customers' transaction histories and preferencesIn asset custody and reconciliation scenarios, its data processing capabilities can automate the parsing and comparison of transaction and valuation information, leading to increased accuracy and reduced workload for bank staff.

Liu Tong highlights that the widespread adoption of DeepSeek fosters a reduction in the "Matthew Effect" within the banking ecosystemThrough the application of DeepSeek technology, banks can achieve intelligent solutions across numerous scenarios while maintaining controlled risksThis enables smaller banks to close the technological gap with larger institutions, while the latter can continue to develop a more robust financial technology ecosystem.

However, as an AI decision-making system, questions about DeepSeek's accuracy in risk assessment persistLiu asserts that ensuring this accuracy involves addressing three primary layers: data, model, and technologyHe also discusses the potential necessity for third-party audit mechanisms to prevent model biases or obscure decision-making

Advertisements

He points out that DeepSeek’s open-source framework inherently improves transparency, allowing banks and related institutions to inspect and validate the model's code and training methodologies, which can partially mitigate some of the risks associated with third-party audit dependencies.

As promising as the applications of DeepSeek may be, Liu cautions that the associated risks cannot be overlookedHe identifies three key areas of concern for banks: model attacks, algorithmic biases, and the leakage of sensitive dataIn response to the risk of model attacks, banks can enhance the robustness of their models through techniques such as adversarial training and data augmentationLiu notes, "While attackers could forge credit materials to deceive the system, integrating results from multiple models can significantly reduce false positives."

Algorithmic bias presents legal and reputational dangersLiu emphasizes the importance of diversified data collection, algorithmic fairness constraints, and dynamic feedback mechanisms to address these biasesRegarding the threat of sensitive data leaks, he explains that while localizing DeepSeek reduces the risk of data outflow, the necessity of accessing large volumes of user data during model training and inference remains an issueMishandling this information could lead to privacy violations, damaging user trust and invoking legal compliance repercussions.

In recent years, specific requirements have been articulated to regulate data processing activities and protect personal informationAs part of this effort, new guidelines were set to take effect by December 27, 2024, stipulating that banks and insurance institutions using AI technologies must explain and disclose the impact of data on decision-making outcomes, regularly monitor automated processes, and design risk mitigation strategies around AI findings.

In the face of these data-driven challenges, Liu advocates that banks utilizing DeepSeek to process sensitive financial information should adopt measures such as data anonymization, access control technologies, and data monitoring and auditing techniques to ensure compliance with data privacy requirements.

On the topic of data anonymization, Liu outlines two key approaches: first, transform sensitive data using specific encoding rules that require unique algorithms to revert; and second, encrypt sensitive financial information—such as transaction passwords and identification numbers—requiring authorized users to use encryption keys for decryption.

When it comes to access control technologies, a combination of user IDs, passwords, digital certificates, dynamic passcodes, and biometric recognition should be employed to ensure that only authorized individuals can access sensitive financial data

Advertisements

Advertisements

Advertisements