Will existing defenses be breached by AI hacking? Financial sector security ‘warning light’

While the possibility that artificial intelligence (AI)-based tools were used in recent hacking attacks targeting the domestic financial sector was raised, an analysis showed that it is difficult to respond to attacks using AI with existing security methods that block specific strings or attack patterns.

Because AI changes attack methods in real time according to the server's response and combines existing web vulnerability attack techniques, there are limitations in finding and blocking traces of specific tools.

In the process of tracking about 50 IP addresses related to attacks on the financial sector, security companies are additionally capturing past intrusion history and signs of prior reconnaissance, leading to criticism that the financial sector's security response system needs to be strengthened.

◇ AI changes attack method in real time… String/header blocking limit

On the 8th, AI Labs, the threat intelligence organization of Monitor Lab, a company specializing in app security, pointed out the limitations of the fragmented hacking response currently taking place in the financial sector in a report on 'Financial sector cyber attacks and AI autonomous mock-hacking tool ARTEX'.

When a Chinese administrator string called 'ARTEX-AI Autonomous Penetration Test Console' was discovered on an attacker's infrastructure server that was presumed to have been used in a financial sector breach, security measures were taken to block the string, specific user agent header, and ARTEX-specific patterns in the firewall.

However, AI Labs defined user agent-based blocking and Artex-specific pattern creation as representative fragmented responses with low effectiveness.

This is because attackers can easily modify header information, and the attack payload (execution code) created by Artex itself is not a new technique but a combination of existing web vulnerability attacks.

According to the report, Artex is not an independent AI model, but a multi-agent-based open source framework that connects existing commercial LLMs such as Claude, ChatGPT, and DeepSeek as a backend.

AI agents autonomously perform the entire process, including preliminary reconnaissance, vulnerability scanning, exploit execution, and bypass attempts in case of failure, which were previously performed manually by hackers.

Ultimately, AI Labs explains that the individual web requests (HTTP) sent by Artex are similar to existing web vulnerability attacks, making it difficult to accurately identify and block the tool with unique signatures alone.

However, AI Labs said that although it is true that traces of ARTEX execution were found in the attack infrastructure, official confirmation is needed to determine whether it was directly executed and used in actual attacks on the financial sector.

◇ Tracking of approximately 50 attacker IPs… Additional traces of intrusion and prior reconnaissance detected

AI Labs announced that it had secured about 50 attacker IP addresses and command and control (C2) server domains related to this financial sector attack using its own threat intelligence and conducted a thorough investigation of security equipment, and additionally detected signs of actual attacks, such as past intrusion history and prior reconnaissance, in some equipment.

AI Labs has urgently reflected and distributed approximately 50 identified attacker IPs and C2 domains in detection patterns.

AI Labs emphasized, "It is difficult to block automated AI attacks by simply blocking one specific string," and added, "We must apply strict authentication and authorization verification to business APIs that are easily left in management blind spots, and combine them with a behavior-based defense system that identifies attack flows and correlations to increase response speed and accuracy."