IN THIS ARTICLE, YOU WILL LEARN:
- Study Overview: NFTEvening and Storible collaborated on a study to test whether AI, specifically Long Short-Term Memory (LSTM) networks, can recover missing cryptocurrency seed phrases by analyzing 85.7 million combinations using the BIP39 word list.
- AI’s Capabilities: The neural network, trained for 30 days, achieved a prediction rate of 994,051 guesses per second. It could recover one missing word in 0.02 seconds, two words in 29 seconds, and three words in 2.28 hours. Recovery of up to four words required 178 hours, with recovery times increasing exponentially for more missing words.
- LSTM Functionality: LSTMs are a type of RNN that excels in sequential data processing, using memory cells and gates to predict and complete patterns in text-like data, making them suitable for seed phrase prediction.
- Security of Crypto Seed Phrases: Despite AI’s advancements, the massive number of possible seed phrase combinations makes it practically impossible for AI to recover enough words to breach cryptocurrency wallets. For example, cracking eight missing words would take 174 times the universe’s age.
- Conclusion and Disclaimer: The study shows that AI is not a current threat to cryptocurrency seed phrase security if the phrase remains secret. However, these findings are based on current AI capabilities and computational limits.
NFT-focused news website NFTEvening and the NFT market’s data and analytics-based platform Storible collaborated to conduct a study to investigate whether powerful AI, specifically a type called Long Short-Term Memory (LSTM), can crack cryptocurrency seed phrases.
For this research, they analysed 85,714,285 seed phrase combinations using 2048 words from the BIP39 word list to determine the time it takes for AI to recover missing crypto seed phrases and the feasibility of LSTM networks in recovering lost or incomplete cryptocurrency phrases.
For this, a neural network was trained on a cloud-based server for 30 days to predict seed phrases and estimate recovery time. The model used LSTM architecture, achieving an average prediction rate of 994,051 guesses per second. The recovery time was analyzed using scenarios from one to twelve missing words.
The results showed that AI could retrieve one missing word in 0.02 seconds, two words in 29 seconds, three words in 2.28 hours, and up to four words in a max recovery time of 178. Moreover, AI can discover the correct seed phrase sequence of 12 words in just 8 minutes, whereas the time needed to recover 8 missing words from a seed phrase is 174 times longer than the universe’s current age.
What is LSTM?
For your information, a seed phrase is like a secret password for your cryptocurrency. It is a list of 12-24 words that gives you access to your digital wallet.
LSTM networks are Recurrent Neural Networks (RNN) that excel in remembering information over extended sequences. They incorporate memory cells and gates to control information flow, allowing them to effectively learn and predict patterns in sequential data like text and time series.
How LSTMs Work in Seed Phrase Recovery
LSTMs can predict the next likely word in a sequence or complete missing words by observing patterns in known seed phrases. Memory cells store crucial information, with three gates – forget, input, and output – regulating the flow of information. The forget gate discards irrelevant information, the input gate determines new information, and the output gate controls the next word prediction.
The researchers tried to use this AI to guess missing words in seed phrases. They tested it on millions of possible combinations. They found that even with this powerful AI, it would take an incredibly long time, longer than the universe’s age, to guess enough words to steal someone’s cryptocurrency.
The research indicates that AI is not a threat to your cryptocurrency if you keep your seed phrase secret. While AI may eventually crack seed phrases, it is currently highly secure. Though LSTMs offer powerful capabilities for sequence learning, the sheer number of possible combinations and the inherent complexity of seed phrase structures make successful recovery highly improbable.
Disclaimer: This analysis provides an estimate based on current AI capabilities and computational power.
