FOMOCheck: AI Agent for Real-Time Detection of Pump-and-Dump Schemes in Solana-Based Decentralized Finance

Abstract

This paper introduces FOMOCheck, an artificial intelligence (AI) agent designed for the early detection of pump-and-dump (P&D) activities in the Solana blockchain ecosystem. The agent integrates CoinGecko market indicators with on-chain token-control features obtained via the Solana API to generate interpretable risk scores and explanations. In a backtesting evaluation involving 30 tokens identified as P&D schemes, FOMOCheck utilized only pre-dump information, achieving a recall of 70.0% and an F1-score of 82.4 % at a risk threshold of 6/10 or higher. A pilot deployment on 20 new token listings demonstrated end-to-end response times under 4.2 seconds (95th percentile), validating its feasibility for real-time triage and monitoring. A qualitative assessment revealed that the system generated transparent, metric-referenced explanations, although it demonstrated sensitivity to “green-flag” cues, such as price-volume alignment. The primary limitations include incomplete data on holder distribution and liquidity locks, as well as the lack of a balanced evaluation set containing non-P&D tokens. Future work will focus on expanding the dataset to include legitimate tokens, integrating streaming data connectors, and deploying a production-ready anomaly scoring API.