South Korea deploys real-time AI crypto surveillance

South Korea’s Financial Supervisory Service has deployed a real-time AI system that scans trading data, news and online content to flag suspected crypto price manipulation.

AI crypto surveillance screens price and volume spikes

The Financial Supervisory Service said in its Aug. 20 announcement that the platform combines generative AI with machine learning to automate parts of a process that previously required investigators to examine large volumes of exchange data manually.

Built around real-time trading information, the system first searches for assets showing abnormal changes in price or volume. It then compares the activity with patterns drawn from the regulator’s previous investigations, allowing staff to focus on trades that share features with known forms of market abuse.

Among the patterns listed by the FSS are the “racehorse” type, in which a token moves sharply during a short period, and the “cage” type, which involves a steep rise in an asset while deposits or withdrawals are suspended or restricted.

The latest platform extends an algorithm introduced in January, when the regulator began using AI to identify suspected price manipulators and isolate the periods and orders linked to their activity. Rather than limiting the technology to a later investigative step, the new setup connects the initial alert, supporting information, and preliminary review in one workflow.

For possible wash trading or coordinated trading, the FSS applies Benford’s Law alongside machine-learning models. Benford’s Law measures how often different leading digits occur in naturally formed numerical datasets, while the regulator uses deviations to select assets and trading periods for further review.

News and chat-room scans test each alert

Once a token records an unusual move, generative AI checks relevant news and exchange announcements for a plausible cause. A listing notice, network update, or another verified event may explain the volatility, while a sharp move without a clear reason can lead the regulator to request detailed order and account data from the exchange involved.

Alongside public market information, the system reviews complaints, tips, and media reports when deciding whether an alert warrants an in-depth analysis. Generative AI then places its findings into a standard report, giving investigators a record of the price move, volume change, identified catalyst, and other indicators before they choose the next step.

Online promotion has also entered the surveillance process. According to the FSS, the system converts text, video subtitles, and audio from YouTube, internet forums, and private-messaging chat rooms into text, then examines the material for suspected front-running, false information, or coordinated calls intended to induce unfair trades.

The FSS said the online review targets cases in which organizers trade ahead of their followers, circulate false claims, or coordinate buy recommendations intended to draw retail traders into an asset. Investigators remain responsible for deciding whether the information supports further analysis or a planned investigation.

South Korea has pursued more than 40 trading cases

The new system follows two years of enforcement under South Korea’s Virtual Asset User Protection Act, which took effect on July 19, 2024. The law requires service providers to separate customer holdings from company assets and keep user deposits with banks, while giving regulators authority to inspect providers and act against insider trading, wash trading, and price manipulation.

As crypto.news reported last month, Korean authorities examined more than 40 suspected unfair-trading cases during the law’s first two years. Financial Services Commission Chair Lee Eog-won said officials reported or referred more than 30 cases to investigative agencies, identified 25 suspects and calculated average unlawful gains of about 1.4 billion won, or roughly $940,000, per case.

Exchange-level controls have developed alongside the regulator’s own surveillance. In May, new API-key controls required members of the Digital Asset Exchange Alliance—Upbit, Bithumb, Coinone, Korbit and Gopax—to monitor suspected key sharing, use IP whitelists and invalidate keys after warnings and user checks.

The rules followed an FSS estimate that API-based trading represented about 30% of domestic crypto turnover. Because an API key can allow an outside program to check balances, place orders, and initiate transfers, the exchange group linked improper sharing to risks that include coordinated trading and possible price manipulation.

Legislation under preparation would cover more than unfair trading. On July 29, the FSC outlined a consolidated bill that could combine 10 pending digital-asset proposals and set rules for stablecoins, exchanges, disclosures, internal controls and system resilience. The Virtual Asset User Protection Act remains the main law governing custody, market abuse and user safeguards while lawmakers negotiate the second-stage framework.

U.S. regulators also keep people in control

In the United States, a May 2025 GAO review found that federal financial regulators used AI to identify risks, support research and detect possible legal violations or reporting errors, but most agencies did not treat model output as the sole basis for a decision.

The Securities and Exchange Commission told the Government Accountability Office that staff used AI tools to identify trading patterns that might indicate insider trading. Subject-matter specialists reviewed the flagged trades before deciding whether further investigation was warranted, while every regulator using AI as of December 2024 said human staff considered model results together with other supervisory information.

At the time covered by the GAO review, federal regulators said they were not using generative AI for supervisory or market-oversight work, although some agencies were considering it. Based on the uses disclosed to the GAO as of December 2024, the Korean platform applies generative AI to supervisory tasks that U.S. agencies had not reported using it for at the time.

A 2025 CFTC roundtable identified real-time detection of spoofing and wash trading as potential uses for AI surveillance. Participants also warned that crypto oversight faces fragmented data because centralized exchanges may execute trades, match orders, manage margin, and hold customer records away from public blockchains.

Under current U.S. law, the Commodity Futures Trading Commission can pursue fraud and manipulation in spot commodity transactions, but it does not routinely supervise spot crypto exchanges in the same way that it oversees registered derivatives markets. Under current CFTC plans, proposals within its existing authority can proceed, while the CLARITY Act would be needed to establish the complete federal registration framework contemplated for spot digital-commodity platforms.

In recent comments, XYO co-founder Markus Levin said regulators need reliable input data and clear operating limits when AI findings can trigger government inquiries. He also raised the risk of false alerts or unverified allegations if investigators place too much weight on automated output.

Levin cited safety tests involving experimental models from Meta, Anthropic and OpenAI that reportedly crossed preset boundaries, accessed systems without authorization or continued operating after restrictions. His comments presented the tests as a warning against allowing automated findings to trigger legal action without independent checks.

Human review will remain mandatory under the FSS process, with investigators assessing each generated report before choosing whether to conduct a detailed analysis or prepare a formal investigation. An FSS official said the platform would help limited staff “respond quickly and efficiently” to increasingly complex unfair trading.

The regulator also plans to add tools for tracing funds across exchanges and following transactions on-chain, although its Aug. 20 announcement did not provide a deployment date for either feature.