AI Can Now Spot Dying Crypto — But Don't Bet Your Savings Yet

Crypto has a graveyard problem. More than 52 percent of all tokens launched since 2021 had stopped trading by early 2025 — meaning millions of people watched their money quietly vanish. Two researchers wanted to know a simple question: can we tell which coins are dying before they're gone? The answer, according to their new paper, is a qualified yes.

They used a type of AI called an LSTM (a model that learns from patterns over time, like reading a story rather than looking at one snapshot). They trained it on 90 days of daily prices and estimated market value for 82 cryptocurrencies — 41 still alive, 41 already dead — pulled from crypto data firm Coin Metrics between 2020 and 2026. To keep the test fair, they only let the AI learn from older data and tested it on newer data, never letting it peek ahead.

On that test set, the results looked stunning: the model correctly flagged failing coins about 98 percent of the time. But when the researchers pointed it at genuinely new, unseen coins, accuracy fell to roughly 59–65 percent — barely better than a coin toss. That gap is the real story. It shows dying coins do leave a fingerprint (slow, steady value loss plus wild price swings in the months before death, rather than a sudden crash), but that fingerprint fades on coins the model has never met before.

So what should you take from this? First, it's early research, not an app you can download — the authors even scrapped a fancier multi-stage version because the data wasn't good enough. Second, the practical warning sign is worth remembering: if a coin's price keeps sliding for months while jumping around unpredictably, that's often a slow death, not a buying opportunity. The paper was accepted to the 2026 International Conference on Blockchain Computing and Applications, where other researchers will try to sharpen it.