KakaoBank said its research on financial AI security and safety has been recognised at an international academic conference. As the use of generative AI spreads across the financial sector, it said it presented research with strong potential for practical application, including detecting malicious content injection attacks, verifying financial calculation errors and evaluating AI safety.

KakaoBank said on Monday that four AI research papers from its Financial Technology Research Institute were accepted in the first half of this year at international academic conferences including ICLR, ACL and LREC.

KakaoBank said it presented a prompt injection detection technology targeting generative AI in specialised fields such as finance and law at ICLR 2026 in April. It uses a dataset of about 59,000 cases jointly built by KakaoBank and KAIST to detect malicious content injection and bypass attacks.

At LREC 2026 held in May, it presented two studies aimed at improving the security and accuracy of finance-focused AI. They include a prompt attack detection technology and a technology that identifies numerical calculation errors that could occur in complex financial data processing.

A study on evaluating the safety of financial AI, conducted jointly with KAIST, was accepted for the ACL 2026 industry track. It proposed AI safety evaluation criteria that classify risk types that may occur in finance, including voice phishing, financial fraud and personal data theft. It also developed a technology to train financial AI models to avoid risky answers.

KakaoBank said, "This research is a practical technology that goes beyond academic achievements to improve the safety and accuracy of financial AI services," and added, "We will continue AI security and technology research specialised for the financial environment."