The Financial Supervisory Service upgraded its AI surveillance platform, VISTA, in early 2026, adding high-performance Nvidia H100 GPUs and new analytics to run near real-time checks across virtual-asset trading. The stated objective is to move supervision from reactive investigations to automated detection after suspicious transactions in 2025 exceeded the prior two years combined.
VISTA now pairs accelerated hardware with a sliding-window grid-search approach designed to flag both short- and long-duration anomalies, from fragmented wash trades to coordinated pump-and-dump behavior. The FSS positioned the upgrade as a capability step-change, saying the system can immediately detect both the number of price-rigging instances and their duration.
How the New Analytics Finds Subtle Manipulation
The core shift is algorithmic: VISTA applies a sliding-window grid search that evaluates every sub-period inside a trading dataset, from seconds to months, and computes abnormal indicators across each interval. By scanning every time slice instead of only fixed windows, the model is built to surface manipulation that is intentionally time-fragmented or subtle enough to evade standard statistical tests.
That analytical intensity depends on higher throughput, which is where the hardware refresh comes in. VISTA now runs on Nvidia H100 GPUs to support the compute demands of sliding-window analysis, and the FSS allocated 117.640 for additional H100 units to expand real-time capacity through 2026. In operational terms, the GPU upgrade is the enabler that turns a computationally heavy detection method into an always-on surveillance workflow.
The upgraded detection set targets wash trading, coordinated account clusters, abrupt volume spikes followed by sharp reversals, and other anomalies that look consistent with price-rigging. VISTA also visualizes trading patterns to help investigators isolate suspicious timeframes with less manual effort. This combination of automated flagging plus visualization is designed to compress case triage time while improving the consistency of how suspicious segments are identified.
What Comes Next and What It Means for Market Participants
The FSS is building additional layers, including a large language model intended to analyze messaging that could indicate collusion, and a separate AI designed to monitor live market trends and distinguish exchange technical glitches from deliberate manipulation. The agency already receives market-trend data daily and plans these tools to support more immediate automated verification. The roadmap signals a shift toward end-to-end surveillance, where behavioral indicators and market microstructure anomalies are evaluated together.
This tighter monitoring posture creates clear second-order effects for the market. Faster automated flags can accelerate inquiries and raise deterrence by increasing the operational difficulty of organized manipulation, but they can also increase false positives and trigger temporary transaction or payment suspensions as a preventive step. The control trade-off is straightforward: broader surveillance coverage can improve enforcement efficiency, yet it can also introduce friction when automated alerts over-trigger.
The upgrade was driven by market context, including notable incidents and the sharp increase in suspicious activity in 2025, which pushed the FSS away from manual, retrospective reviews and toward continuous surveillance. By emphasizing quantitative signals like both the count and duration of rigging episodes, the agency is building a more metrics-driven enforcement posture.
For traders, corporate treasuries, custodians, and exchanges, the operating implications are practical rather than theoretical: surveillance sensitivity is rising, abnormal order flow will face tighter scrutiny, and flagged activity may lead to faster outreach or temporary freezes. Firms that tighten internal monitoring rules, liquidity controls, and counterparty screening will be better positioned to avoid false flags and respond quickly to regulatory requests.
VISTA will continue expanding through 2026 as additional H100 units are installed and the planned LLM and real-time monitoring tools roll out. As that footprint grows, automated detection will become more comprehensive and faster, changing execution conditions for volatile tokens and derivatives while giving regulators more granular enforcement signals.