Beyond Core Banking: How Agentic AI Exposed Soft Spots in South Korea’s Financial Sector

Beyond Core Banking: How Agentic AI Exposed Soft Spots in South Korea’s Financial Sector

Cybercriminals infiltrated seven major South Korean financial institutions, exfiltrating more than 67,000 sensitive customer records.

Technical intelligence examining the campaign’s infrastructure reveals how a financially motivated, Chinese-speaking threat actor deployed an agentic artificial intelligence framework to autonomously discover vulnerabilities, bypass authentication and map attack paths across auxiliary systems.

Instead of launching a direct assault against heavily guarded core transactional networks, the attacker weaponized a newly released, open-source multi-agent autonomous penetration-testing tool known as ARTEX.

Operating across a two-server infrastructure anchored by a Hong Kong backbone IP address, the threat actor linked the ARTEX engine directly to multiple Large Language Model  backends via API keys to execute automated reconnaissance and exploitation.

The ARTEX framework relied on DeepSeek V4.1-Flash as its primary reasoning engine to execute core automated scanning workflows. To refine reconnaissance, the operator supplemented scanning tasks using Zhipu AI’s GLM-5.3 and SpaceXAI’s Grok 4.6.

During post-exploitation phases, the threat actor utilized Anthropic’s Claude Code to analyze target system responses and query regional Telegram channels to locate buyers for the stolen financial data.

By chaining LLM reasoning with automated pentesting scripts, the agentic framework systematically probed internet-facing web applications without requiring constant human intervention.

The AI scanner identified unmonitored web interfaces and soft spots in secondary support platforms rather than primary banking systems.

At one bank, the attacker leveraged the agentic tool to bypass authentication on a loan-progress inquiry service used by third-party financial brokers, exposing 25,000 customer records. At another institution, the agent breached an employee mobile work-support portal.

The attacker successfully extracted customer personal details and loan files across affected entities including Shinhan Bank, KB Kookmin Bank, Hana Bank, BNK Busan Bank, Yegaram Savings Bank, Welcome Savings Bank and Hyundai Capital. Two additional institutions, Woori Bank and NH NongHyup Bank, detected automated intrusion attempts but intercepted the attack vectors before data exfiltration occurred.

To counter AI-driven reconnaissance engines, enterprise defenders must modernize their perimeter controls.

Security teams need to deploy inline IPS and encrypted traffic inspection to catch automated model-generated vulnerability probing, while enforcing strict Zero Trust architecture and multi-factor authentication across all third-party broker and employee support portals.

Additionally, organizations must apply strict rate limiting on data-lookup APIs, monitor outbound API calls to unauthorized LLM endpoints to eliminate Shadow AI, and maintain continuous attack surface management to discover and patch exposed web applications before autonomous AI agents index them.

Senior Associate Editor - APAC (CISONow)
Senior Associate Editor at  | Editorial | Bilaspur, Chhattisgarh | prajeet@observenow.co.in | + posts

Nair is a seasoned cybersecurity journalist with over a decade of experience covering cybersecurity and OT developments in the US and the Asia-Pacific region. As an editor, he has interviewed key decision-makers, including CISOs, CIOs, regulators and law enforcement leaders. Before joining ObserveNow, Nair held editorial roles at Information Security Media Group, News Corp owned TechCircle, IDG, The Times Group and The New Indian Express. He is currently based in Bilaspur, India.

Share:

Author

Prajeet Nair

Nair is a seasoned cybersecurity journalist with over a decade of experience covering cybersecurity and OT developments in the US and the Asia-Pacific region. As an editor, he has interviewed key decision-makers, including CISOs, CIOs, regulators and law enforcement leaders. Before joining ObserveNow, Nair held editorial roles at Information Security Media Group, News Corp owned TechCircle, IDG, The Times Group and The New Indian Express. He is currently based in Bilaspur, India.

Chat with CISONow.in