AIBRIEF
Researchers tracked 6.4 million memecoins — a type of crypto token — on Solana, a blockchain network, and found most exhibited a 'rug pull,' a liquidity drain scam, within an hour of launch.
Most new Solana memecoins are not projects that later turn out to be scams. They are scams that briefly pretend to be projects, and a new arXiv preprint says a machine-learning model can spot the difference within the first five minutes of trading.
That's faster than the hour it usually takes the price to collapse, but not fast enough to save the buyers who showed up during those same five minutes.
Researchers tracked 6.4 million tokens issued on the Solana blockchain, a network favored for cheap, fast token launches, over seven months. A large share of them showed "rug pull" behavior, a pattern where the creators withdraw the trading pool and the price collapses to zero, within an hour of going live.
The authors fed only the first five minutes of trading data into gradient boosting, a widely used family of decision-tree models. With no code-level signals to work from, the model flagged a meaningful share of imminent rug pulls, the authors report. Adding data from Raydium, the decentralized exchange where the tokens trade, made detection noticeably more reliable than signals from PumpFun, the launchpad, alone.
The structural limit is in the clock. A five-minute warning arrives after the first wave of retail buyers has already handed over their money, and the paper's own emphasis on cross-platform fusion, not the model alone, makes the same point. The preprint, arXiv 2608.20271, is not peer-reviewed, and the headline numbers should be read as the authors' own report.