South Korea's financial regulator will begin full operation in 2028 of an artificial intelligence model that detects and investigates unfair trading in capital markets, including stock price manipulation. The AI will identify abnormal trades and linked accounts, and propose the additional evidence needed for further investigation. LG CNS (064400), the IT services affiliate of LG Group, will develop the system.
The project, worth about 4.5 billion won (roughly $3.2 million), was awarded to LG CNS after it competed with Bespin Global and Konan Technology (402030) for preferred bidder status. The system is scheduled for completion by December 2026, with full deployment the following year.
How the AI Investigation System Works
The core task is using AI to flag suspected unfair trading early - including leveraged buyouts financed without capital and stock price manipulation - and to improve investigative efficiency. LG CNS plans to build the model by fine-tuning an existing public general-purpose AI model for the FSS's specific work.
The model will learn from public information such as share prices, trading volumes and regulatory filings, plus non-public material the FSS has obtained during past investigations. Historical case records at the FSS, including investigation plans, reports and disposition opinions, will serve as key training data.
On that basis, the AI will detect trades suspected of unfair trading and automatically map relationships among multiple accounts. It will group suspicious accounts for analysis of trading circumstances and suggest to investigators what additional material is needed to establish charges. When FSS staff collect that material and feed it back into the system, the AI reanalyzes the likelihood of violations based on the new input - a circular process meant to sharpen investigative precision over time.
Closed Network and AI Agent Support
Because the system handles sensitive investigative information, the AI model will be installed on a closed network separated from the external internet. The FSS barred external cloud-based AI services such as OpenAI's ChatGPT or Google's Gemini for this project. LG CNS's proposal is understood to include two domestic large language models as candidates, among them LG AI Research's EXAONE, which has publicly available model weights and can run on in-house servers.
The project also includes an AI agent to assist FSS staff, provided primarily in chatbot form. When a user asks in natural language about suspected unfair trading at the current moment, the agent responds based on real-time data. It will also draft reports such as question-and-answer records and disposition opinions in FSS formats. All AI agents will be embedded in the interface of the FSS's existing work systems, alongside tools to assess the priority of tips and complaints and automatically transcribe and summarize recordings.
"We will mobilize all of our expertise in financial services along with our AI model, platform and IT service capabilities to build an AI that contributes to innovating the FSS's response to unfair trading," an LG CNS official said.
Why this matters for finance professionals
The system doesn't replace human investigators - it changes what they do. For compliance, risk and data professionals at financial institutions, this means more advanced surveillance of trading patterns across large account networks. The AI's ability to trace linked accounts and suggest evidence will likely lead to faster, more targeted inquiries into broader market participants, including institutional trading. Those working in finance and trading will want to understand what patterns the FSS's AI flags, since investigation standards and nine-digit won penalties are will only tighten over time. For professionals in AI for Finance, this project shows how a regulator applies fine-tuned Generative AI and LLM tools to close monitoring of market activity - a clear signal that similar systems will soon be adopted by private financial firms globally.
Your membership also unlocks:
