Selectstar logo.

Selectstar, an AI data and reliability evaluation company, said on Monday a research paper on K-BankEval, a finance-focused large language model (LLM) evaluation benchmark it jointly developed with Shinhan Bank, was accepted for a workshop at the international academic conference ICAIF 2026.

ICAIF 2026 is an international academic conference in the field of financial AI held in Milan, Italy from Nov. 14. Researchers from the two companies presented their results at a workshop on financial AI security, safety and privacy held on Nov. 15.

K-BankEval is an evaluation framework to select AI models suitable for real banking work. It consists of a total of 5,000 evaluation items across six areas: financial knowledge, retrieval-augmented generation (RAG), tool calling, financial reasoning, answer safety and AI agent task planning.

It evaluates answer safety for financial consumer protection and AI agents' ability to perform tasks, supplementing the limitations of existing general-purpose LLM evaluation methods. Shinhan Bank presented the evaluation criteria needed for actual work, and Selectstar was in charge of developing the evaluation technology.

Selectstar has previously had a paper on FinRED, a financial AI safety evaluation framework jointly developed with the Financial Security Institute and the Korea Advanced Institute of Science and Technology (KAIST), accepted for the international conference IEEE ICDM 2026. FinRED is a technology that verifies AI safety based on financial risk situations such as voice phishing and money laundering.

Selectstar CEO Seyop Kim (김세엽) said, "We have established a framework that concretely defines the AI capabilities required in real banking work and can evaluate them objectively." He added, "We will strengthen our technological capabilities to verify the job suitability and safety of financial AI."

Keyword

#Selectstar #Shinhan Bank #K-BankEval #ICAIF 2026 #FinRED
Copyright © DigitalToday. All rights reserved. Unauthorized reproduction and redistribution are prohibited.