Financial Security Institute establishes security evaluation standards for AI agents in the financial sector

17 items prepared… Significantly expands AI agent security

(Seoul = Yonhap News) Reporter Kang Soo-ryeon = The Financial Security Institute announced on the 7th that it has prepared 'Financial AI Agent Security Evaluation Standards' to check new security threats due to the expansion of artificial intelligence (AI) agents in the financial sector.

An AI agent is an artificial intelligence system that makes plans on its own to achieve a given goal and performs actual tasks using tools and external systems.

Since it autonomously handles everything from problem definition to execution without human intervention, security threats are more likely to affect actual work than existing AI.

The Financial Security Institute decided to analyze major domestic and foreign AI agent security threats and response measures to derive inspection items and reflect them in the 'AI Redtiming' standards.

AI Red Teaming is a mock hacking technique that attacks in advance from a hacker's perspective, discovers vulnerabilities, and checks the adequacy of the response.

The evaluation criteria are divided into 17 items in 5 areas, including ▲data contamination attack response ▲model contamination attack response ▲data/model extraction attack response ▲evasion attack response ▲AI agent security.

Among these, there are 10 items to prepare for new threats caused by the introduction of AI agents.

Confirmed AI agent security threats, actual attack techniques, and final evaluation criteria will be disclosed in this year's 'AI Red Team Report'.

Director Park Sang-won said, “With the establishment of this evaluation standard, we will substantially verify the risks of AI agents in the financial sector and actively support financial companies’ use of safe and reliable AI technology.”