Kbank has adopted a large language model (LLM) on its path to becoming an artificial intelligence (AI)-powered bank, the internet-only lender said Wednesday.

It is the first online-only bank in Korea to integrate an LLM for its business at a time when AI is accelerating digital transformation in the finance sector.

An LLM is an AI system capable of understanding and generating human language by processing vast amounts of advanced statistics and data.

Headquarters of internet-only lender Kbank in central Seoul / Courtesy of Kbank

Kbank said its LLM is customized as a private model, meaning it keeps sensitive data within the company and prevents data from being exposed to outside parties.

The announcement came as Kbank has been pioneering paths to fully implement non face-to-face banking services after it was established as the country's first online lender in 2016.

The banking arm of telecom giant KT, the firm said the LLM was made possible through collaboration with KT and two other stakeholders — KT Cloud and Upstage — under a joint memorandum of understanding signed in February 2024.

Kbank underlined that its model has an advantage over open LLMs, given that a private LLM allows companies to train the model with their desired specialized data, resulting in more accurate and reliable information generation.

It also said its model offers heightened security as the data is not shared externally and is processed only on internal servers.

“Privacy is especially crucial in the financial sector, which involves complex terminology and concepts and places a high emphasis on security. As such, the introduction of a private LLM with specialized performance and robust security is gaining significant attention,” Kbank said.

The lender also said it gathered “a vast amount of data" — about 100 million books from financial institutions, academia and other relevant groups — to develop its LLM.

The model was fine-tuned with financial data to enhance its expertise in the financial domain while maintaining its general language processing capabilities.

Such efforts have improved the model's accuracy in answering finance-related questions and increased its reliability.

The model was additionally evaluated using data from more than 20 financial certification exams, to measure its financial knowledge level and overall performance.