Concerns that AI-based autonomous penetration tools may have been deployed in cyberattacks targeting South Korea’s financial sector are exposing the limits of existing security systems. Analysts say that because attackers use AI to change their tactics in real time, traditional methods of blocking specific strings or attack patterns are hard-pressed to keep up.
AI Labs, the threat intelligence organization of app security specialist MONITORAPP, reached that conclusion in an August 8 report titled “Cyberattacks on the Financial Sector and the AI Autonomous Penetration-Testing Tool ARTEX.” After the Chinese-language administrator string “ARTEX-AI Autonomous Penetration Testing Console” was found on an attacker infrastructure server suspected of involvement in a recent financial-sector breach, security teams moved to block that string, a specific user-agent (UA) header and ARTEX-specific patterns at their firewalls.
AI Labs, however, characterized such responses as a textbook example of low-effectiveness, piecemeal measures. Attackers can easily forge header information, and the attack payloads (execution code) that ARTEX generates are not new techniques but combinations of existing web vulnerability exploits.
According to the report, ARTEX is not a proprietary AI model but an open-source, multi-agent framework that connects existing commercial large language models (LLMs) such as Claude, ChatGPT and DeepSeek as backends. AI agents autonomously carry out the entire process that hackers once performed manually — preliminary reconnaissance, vulnerability scanning, exploit execution and, when an attempt fails, evasion retries. In the end, the individual HTTP requests ARTEX sends resemble conventional web attacks, making it difficult to precisely identify and block the tool by its unique signature alone.
AI Labs drew a line, saying that while traces of ARTEX execution were indeed found in the attack infrastructure, official confirmation is needed as to whether it was actually run and used in the attacks on the financial sector.
Separately, AI Labs said it used its own threat intelligence to obtain roughly 50 attacker IP addresses and command-and-control (C2) server domains linked to the financial-sector attacks and conducted a full inspection of security equipment, additionally detecting signs of real attacks — including past intrusion histories and preliminary reconnaissance — on some devices. The identified attacker IPs and C2 domains have been urgently incorporated into detection patterns and distributed.
“It is difficult to fend off automated AI attacks simply by blocking a single specific string,” AI Labs said. “We must apply strict authentication and authorization verification even to business APIs that tend to fall into management blind spots, and combine that with a behavior-based defense system that identifies attack flows and correlations to raise both the speed and accuracy of responses.”
The analysis carries significant implications for the direction of the financial sector’s security investments. As existing defenses centered on signature- and string-based blocking struggle to filter out attacks that AI mutates on its own, there is growing support for the view that a shift toward a multi-layered response system — including behavior-based detection and stronger API authentication — is needed. Financial companies are expected to share and incorporate the already-detected attacker infrastructure information while inspecting segments that have been relatively loosely managed, such as internal business APIs and administrator consoles.
Click Here For The Original Source.
——————————————————–
