Algorithms Track ‘Racehorse’ and ‘Cage’ Schemes #
South Korea’s financial regulator has reportedly deployed a real-time, artificial intelligence-powered market surveillance system to detect unfair trading and price manipulation across the country’s cryptocurrency markets. The Financial Supervisory Service (FSS) announced that the new platform combines generative AI with machine-learning algorithms.
The initiative aims to overcome severe resource constraints faced by human investigators, who must monitor thousands of digital assets trading around the clock across multiple exchanges. The system expands on algorithms developed by the regulator in January that pinpointed price-manipulation suspects and the timing of their orders. With this latest rollout, AI oversight now spans the entire end-to-end surveillance pipeline.
According to a local report, a core feature of the system is its capability to flag short-term price manipulation by referencing a historical repository of known market abuse tactics. Among the primary targets are “racehorse” schemes — where traders rapidly inflate token prices over short periods — and “cage” schemes, where assets under temporary deposit and withdrawal limits experience sharp price swings.
To detect volume inflation, the system pairs Benford’s law with machine-learning models to flag wash trading and coordinated group trading. When suspicious volume or price spikes trigger an alert, generative AI automatically scans related news coverage and official exchange announcements to assess whether the market movement stems from legitimate fundamental news. If no valid reason is found and manipulation remains likely, the FSS requests granular trade data directly from the affected exchange.
The agency’s surveillance grid reaches beyond order books and exchange logs. The platform actively monitors digital channels, including Youtube, online message boards, and private messaging app chat rooms. By converting video audio and subtitles into text in real time, the AI evaluates media for signs of front-running, false information dissemination, and coordinated buying advice designed to trap retail investors.
Despite the heavy automation, human oversight remains central to the process. FSS investigators evaluate the AI-generated reports before deciding whether to launch formal enforcement proceedings. While the agency frames the deployment as a necessary step to police complex crypto markets, industry observers have expressed skepticism about relying on AI for regulatory enforcement.
Markus Levin, co-founder of XYO, raised concerns about how regulators can ensure these systems operate on accurate, trustworthy data. He questioned whether AI models can be trusted to remain within clearly defined operational boundaries, particularly when their output directly triggers government investigations or formal legal action.
Levin highlighted recent safety tests conducted by leading AI developers, including Meta, Anthropic and OpenAI. In several instances, experimental models bypassed intended safety boundaries, accessed unauthorized systems, and persisted in actions after encountering operational restrictions. Relying on similar automated models to drive regulatory enforcement, critics argue, introduces significant risks of false positives or unverified claims.
Despite these concerns, the FSS plans to continue upgrading its market oversight toolkit. Upcoming iterations of the platform will focus on integrating tools to trace cross-exchange fund flows and monitor on-chain blockchain transactions, further bolstering the agency’s ability to police digital asset markets.