Banksalad has developed technology to control incorrect financial judgments and transaction execution by AI agents and has filed three related patent applications. The core is a verification process designed to prevent transactions that could disadvantage customers when AI misanalyses financial product information or triggers calculation errors.

Banksalad said on Wednesday it completed filing three patent applications related to its "financial AI agent decision and execution control technology."

The technology focuses on three functions: judging what to execute, verifying figures and controlling execution. It aims to address so-called hallucination errors in which AI generates information that differs from reality.

First, it developed technology that analyses a customer's credit status and loan terms at a preliminary loan screening stage to identify customers who can borrow at a lower interest rate. It improves decision accuracy by training the AI prediction model on actual loan screening results.

It also prepared technology to check for missing information or calculation errors when analysing financial product information and benefits. It is designed to conduct final cross-verification of figures calculated using formulas based on financial rules such as the debt service ratio (DSR) and loan-to-value (LTV) ratio.

The technology also includes verifying the accuracy of input information and figures before an AI agent executes a financial transaction, and deferring execution if the transaction is judged to be disadvantageous to the customer.

Banksalad plans to use the technology to reduce risks that could arise in the financial judgment and execution process of AI agents. The procedure completed this time is at the application stage, not patent registration.

A Banksalad official said, "We take as an important principle not only that financial AI agents provide financial benefits to customers, but also that they verify and control errors so customers can use services safely."