AI picks a million passwords per second to crack cryptocurrency wallets

A study conducted by NFT platforms NFTEvening and Storible analyzed the capabilities of a neural network based on the Long Short-Term Memory (LSTM) model to recover cryptocurrency seed phrases.

Seed Phrases — These are the keys to digital wallets, which consist of 12 minus 24 words, and their security is critical for the safety of cryptoassets (ed.).

In the experiment, 85,714,285 seed-phrase combinations were tested, consisting of 2048 words from the BIP39 list. The main goal was to determine how quickly and efficiently an LSTM neural network can recover lost or incomplete phrases.

LSTM (Long Short-Term Memory) — it is a type of recurrent neural network that specializes in working with sequential data, such as text or time series. This architecture has memory cells and&nsbsp;three types of control «gates» (forgetting, input and output), which allows you to analyze complex patterns and predict missing elements (ed.).

The neural network was trained on a cloud server for 30 days. The average matching speed was 994 051 guesses per second. The recovery time depended on the number of missed words:

• One missed word: 0.02 seconds.
• Two words: 29 seconds.
• Three words: 2.28 hours.
• Four words: up to 178 days.

If part of the seed phrase was known, the neural network could fully recover the 12-word sequence in 8 minutes. In the absence of any initial information, the selection time increased significantly, and the probability of success became extremely small.

The researchers note that, despite the high speed and power of modern artificial intelligence, the number of possible combinations for selecting seed phrases remains enormous, which makes their successful recovery unlikely. Thus, cryptocurrency assets remain safe if seed phrases are kept secret.