South Korea’s FSS Builds AI System for Real-Time Crypto Market Surveillance

South Korea’s Financial Supervisory Service has built an artificial intelligence-based market surveillance system to detect and analyze unfair trading in digital assets in real time, including price manipulation and wash trades.

The Financial Supervisory Service, or FSS, said on Aug. 20 that it designed and built the system in-house by combining generative AI with machine-learning algorithms. The move is aimed at improving oversight efficiency in a market where thousands of tokens trade around the clock across multiple exchanges, from detecting unusual transactions to determining whether an investigation is warranted.

The FSS has gradually expanded the use of AI in digital-asset market surveillance since early 2026. In January, it developed an algorithm that automatically identifies bid orders and trading periods tied to accounts suspected of price manipulation. In April, it added a function that automatically identifies multiple linked accounts used in such schemes. The latest upgrade extends AI across the surveillance process, including real-time monitoring of unusual trades, in-depth analysis and decisions on whether to open investigations.

The new system first screens for tokens showing abnormal price moves over short periods using real-time data from exchanges. It is designed to detect ultra-short-term pump-and-dump patterns, including the so-called racehorse type, in which prices swing sharply during a specific time window, and the pen type, in which prices are pushed up in tokens whose deposits and withdrawals have been suspended or restricted.

When unusual activity is detected, generative AI reviews exchange notices and related news to analyze the drivers of sudden changes in price and volume. If a token surges without a clear positive catalyst, or if typical manipulation patterns are identified, the FSS obtains trading data from the exchange and further reviews whether a formal investigation is needed.

The agency is also using AI to identify wash trades and matched orders. To detect fabricated trades that artificially inflate volume, it applies Benford’s law alongside machine learning. After checking how far the digit distribution in each token’s trading volume deviates from normal patterns, it uses algorithms including autoencoders and isolation forests to flag trading periods that differ sharply from ordinary activity.

The surveillance scope extends beyond exchange data. The FSS is also using AI to analyze illegal front-running conducted through online chat rooms, as well as videos containing false information and online posts that encourage unfair trading.

It collects digital-asset-related posts and videos through application programming interfaces, or APIs, converts audio and subtitles from videos into text, and uses generative AI to assess the potential for violations. The system evaluates risk by distinguishing whether content is intended to induce investor trading, report another person’s unfair trading, or simply reflect casual conversation. It then links that analysis to actual token price moves to examine whether misconduct may have occurred.

The FSS has also automated in-depth analysis after AI detects unusual trades. It selects tokens for review by combining real-time monitoring results with complaints, tips and media reports. The process reflects not only quantitative indicators such as price and volume changes, but also qualitative factors including AI analysis of what drove the price move.

Once a token is selected for in-depth analysis, the agency uses its existing trade-analysis system, generative AI and separate AI algorithms to identify suspicious accounts and suspect trading periods. It analyzes indicators such as each account’s influence on quoted orders, influence on prices and trading profit, then traces linked accounts to identify groups of suspect accounts and the relevant trading windows.

Generative AI also drafts review reports based on the analysis. After AI prepares reports in a set format using various indicators for judging suspected misconduct, investigators review them and decide whether additional analysis or a planned investigation should begin.

The FSS plans to further upgrade the AI surveillance system by adding functions that support fund-flow analysis and on-chain tracking.

The agency said the new AI-based market surveillance system should enable faster and more efficient responses to increasingly sophisticated and complex digital-asset market misconduct despite limited staffing. It added that it will further strengthen its AI-based surveillance and investigative framework to protect users and establish a sound digital-asset market order.