Two ways the house wins: one sells you hype until the price is high enough to dump on you, the other just walks off with the money once enough of it is in the room.
A pump-and-dump works from outside the project: a coordinated group builds artificial hype around a token they already hold, drives the price up on that hype, then sells into the rise — leaving everyone who bought in near the top holding a coin with no support underneath it.
A rug pull works from inside the project: the (often anonymous) team behind a token builds hype, attracts real money into a liquidity pool, then withdraws the funds outright, leaving the token worthless. The difference is who’s doing the extracting — outside manipulators versus the project’s own creators — but the effect on a retail buyer is the same.
Both are common enough on high-speed, low-cost networks like Solana that a large share of newly launched tokens show warning signs of one or the other before they’ve even finished trending. Recent research using machine learning on-chain has found these patterns detectable early — in some cases within the first five minutes of a token’s trading — from signals like liquidity pool behaviour and token distribution alone, before most human traders would notice anything wrong.
Source: Yaremus, D., Li, J., Kalacheva, A., Vodolazov, I., & Yanovich, Y. (2025), “Detecting Rug Pulls in Decentralized Exchanges: Machine Learning Evidence from the TON Blockchain,” arXiv:2509.01168.