Four papers written by KakaoBank's in-house Financial Technology Research Institute on improving the safety of Generative AI and enhancing the reliability of financial AI were consecutively accepted at international academic conferences.
According to KakaoBank on the 30th, the company presented technology to detect harmful content injection and bypass attacks (prompt injection) targeting Generative AI in specialized fields such as finance and law at ICLR in Apr., and then presented two studies that improve the security and accuracy of finance-specialized AI at LREC in May, a leading international conference in natural language processing, language resources, and evaluation.
In the industry track of ACL, research jointly conducted by KakaoBank and KAIST in May was accepted. KakaoBank proposed AI safety evaluation criteria that systematically organize risk types that can occur in finance, including voice phishing, financial fraud, and personal information theft.
KakaoBank said, "It is meaningful in that it is practical technology that improves the safety and accuracy of financial AI services," adding, "We will continue research on AI security and technology tailored to the financial environment to provide the safest and most convenient financial services."