(Editor's Note: This version of the article has been updated to correct a slug and title.)
Chinese banks' embrace of DeepSeek could expose them to risk management challenges. S&P Global Ratings believes that the application of generative AI in the banking sector will generally be evolutionary rather than revolutionary. Nevertheless, aggressive application to credit approvals could bring potential asset quality problems that outweigh operational benefits.
What's Happening
As of February 2025, at least 19 Chinese lenders have sought to transform their business by implementing DeepSeek. These lenders including three megabanks, two joint-stock banks and 14 smaller regional and internet banks (collectively representing 33% of system assets) Use cases start in internal applications but will also be employed in customer-facing ones.
Why It Matters
The use of DeepSeek will likely facilitate efficiency gains. Generative AI is established within Chinese lenders. Some large players began exploring similar applications a few years ago, which we view as being more proactive than counterparts in Europe and U.S. But we expect DeepSeek's lower-cost advantage will widen generative AI access in the banking sector over the next few years, particularly for some resource-constrained small banks. Broader generative AI adoption will likely streamline certain resource-intensive tasks.
But generative AI adoption could also endanger banks' risk controls. Certain types of credit modeling and scorecard risk management have delivered mixed results in consumer and microfinancing. Potential unintended consequences include:
AI-generated misinformation: This could compromise underwriting quality when fake or inaccurate output is use to approve credit. So-called model hallucination is likely pronounced in the analysis of non-standard information such as behavioral data.
Data limitation: Currently, generative AI use reinforces banks' existing credit assessment systems without providing additional client information. At the same time, a small sample size and uneven geographical data distribution may limit comprehensive risk analysis, particularly for smaller banks.
Complexity risk: The somewhat opaque nature of generative AI modeling makes regulatory scrutiny challenging, even with banks' efforts in transparency. This also tests data protection and cybersecurity robustness.
What Comes Next
Much will depend on Chinese banks' digital governance. We anticipate the sector will approach generative AI use with caution. By employing systems and tools to train, validate, and improve accuracy and reliability of their risk-management models banks could alleviate unseasoned risks from model integrity.
Meanwhile, open-source models like DeepSeek can offer an advantage over proprietary models in reducing vendor lock-in risk, provided that banks effectively manage the underlying technology and the development process.
That said, banks that use generative AI to make lending decisions without thorough model validation and risk calibration will likely face higher credit risks and possible business disruption.
We expect regulators will continue to enforce banks' tightening control of core AI capacities to contain potential system risks. For banks with weak governance, regulators may rein in the use of generative AI. Regulators have tightened banks' underwriting requirements in fintech-fueled microfinancing since 2017 mainly to control small banks' retail lending risks. In recent years, digital tools have enabled real-time regulatory monitoring of model-driven credit approval and risk management in Chinese banks on a selective basis.
Background In Brief
The adoption of generative AI comes at a time when Chinese banks are increasingly focused on cost control to maintain profitability amid margin pressure. China's rate cuts to support the economy have affected the banking sector's cost efficiency, with the average cost-to-income ratio rising to 35.6% in 2024 from 31.7% five years ago, although it remains low compared to global standards.
Related Research
- AI Brief: DeepSeek Is A Catalyst For Innovation, Feb. 4, 2025
- Your Three Minutes In AI: Financial Systems Will Face New Systemic Risks, Oct. 4, 2024
Please also see our research landing page covering technological disruption in the sector: The Future of Banking.
This report does not constitute a rating action.
| Primary Credit Analysts: | Michael Huang, Hong Kong + 852 25333541; michael.huang@spglobal.com |
| Robert Xu, Hong Kong + 852 2532 8093; Robert.Xu@spglobal.com | |
| Secondary Contact: | Ryan Tsang, CFA, Hong Kong + 852 2533 3532; ryan.tsang@spglobal.com |
| Research Assistant: | Jiawen Zhang, HANGZHOU |
No content (including ratings, credit-related analyses and data, valuations, model, software, or other application or output therefrom) or any part thereof (Content) may be modified, reverse engineered, reproduced, or distributed in any form by any means, or stored in a database or retrieval system, without the prior written permission of Standard & Poor’s Financial Services LLC or its affiliates (collectively, S&P). The Content shall not be used for any unlawful or unauthorized purposes. S&P and any third-party providers, as well as their directors, officers, shareholders, employees, or agents (collectively S&P Parties) do not guarantee the accuracy, completeness, timeliness, or availability of the Content. S&P Parties are not responsible for any errors or omissions (negligent or otherwise), regardless of the cause, for the results obtained from the use of the Content, or for the security or maintenance of any data input by the user. The Content is provided on an “as is” basis. S&P PARTIES DISCLAIM ANY AND ALL EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, ANY WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE OR USE, FREEDOM FROM BUGS, SOFTWARE ERRORS OR DEFECTS, THAT THE CONTENT’S FUNCTIONING WILL BE UNINTERRUPTED, OR THAT THE CONTENT WILL OPERATE WITH ANY SOFTWARE OR HARDWARE CONFIGURATION. In no event shall S&P Parties be liable to any party for any direct, indirect, incidental, exemplary, compensatory, punitive, special or consequential damages, costs, expenses, legal fees, or losses (including, without limitation, lost income or lost profits and opportunity costs or losses caused by negligence) in connection with any use of the Content even if advised of the possibility of such damages.
Credit-related and other analyses, including ratings, and statements in the Content are statements of opinion as of the date they are expressed and not statements of fact. S&P’s opinions, analyses, and rating acknowledgment decisions (described below) are not recommendations to purchase, hold, or sell any securities or to make any investment decisions, and do not address the suitability of any security. S&P assumes no obligation to update the Content following publication in any form or format. The Content should not be relied on and is not a substitute for the skill, judgment, and experience of the user, its management, employees, advisors, and/or clients when making investment and other business decisions. S&P does not act as a fiduciary or an investment advisor except where registered as such. While S&P has obtained information from sources it believes to be reliable, S&P does not perform an audit and undertakes no duty of due diligence or independent verification of any information it receives. Rating-related publications may be published for a variety of reasons that are not necessarily dependent on action by rating committees, including, but not limited to, the publication of a periodic update on a credit rating and related analyses.
To the extent that regulatory authorities allow a rating agency to acknowledge in one jurisdiction a rating issued in another jurisdiction for certain regulatory purposes, S&P reserves the right to assign, withdraw, or suspend such acknowledgement at any time and in its sole discretion. S&P Parties disclaim any duty whatsoever arising out of the assignment, withdrawal, or suspension of an acknowledgment as well as any liability for any damage alleged to have been suffered on account thereof.
S&P keeps certain activities of its business units separate from each other in order to preserve the independence and objectivity of their respective activities. As a result, certain business units of S&P may have information that is not available to other S&P business units. S&P has established policies and procedures to maintain the confidentiality of certain nonpublic information received in connection with each analytical process.
S&P may receive compensation for its ratings and certain analyses, normally from issuers or underwriters of securities or from obligors. S&P reserves the right to disseminate its opinions and analyses. S&P's public ratings and analyses are made available on its Web sites, www.spglobal.com/ratings (free of charge), and www.ratingsdirect.com (subscription), and may be distributed through other means, including via S&P publications and third-party redistributors. Additional information about our ratings fees is available at www.spglobal.com/usratingsfees.