HOW FAST CAN AI RECOVER YOUR SEED PHRASE?
We analyzed 85,714,285 seed phrase combinations using 2048 words from BIP39 Word List to estimate how quickly AI could recover lost or incomplete crypto seed phrases.
Click on any word in the seed phrase to burn
YOUR WALLET SEED PHRASE / 20234
*The seed phrase is only for demonstration purposes
Number of missing word in Seed Phrase
0
An AI would need about
to recover your Seed Phrase
Tweet Your Result
WHAT WE DO
We use LSTM networks to predict and fill in missing words based on patterns learned from a set of sample seed phrases
Below is what we foundTIME IT TAKES FOR AI TO RECOVER YOUR SEED PHRASE IN 2024
| No. of Missing Word | Time to Recover | Combinations | |
|---|---|---|---|
| 1 | 0.0204 Seconds (0.0204s) |
 |
2048 |
| 2 | 28.98 Seconds (28.98s) |
 |
2048^2 |
| 3 | 2.28 Hours (8,229s) |
 |
2048^3 |
| 4 | 177.79 Days (15,361,848s) |
 |
2048^4 |
| 5 | 892.82 Years (28,175,256,432s) |
 |
2048^5 |
| 6 | 1.85 Million Years (1,850,427 Years) |
 |
2048^6 |
| 7 | 1.59 Billion Years (1,592,739,726 Years) |
 |
2048^7 |
| 8 | 2.39 Trillion Years (2,391,668,009,669 Years) |
 |
2048^8 |
| 9 | 18.2 Quadrillion Years (18,211,120,064,464,100 Years) |
 |
2048^9 |
| 10 | 36.3 Quintillion Years (36,319,117,647,058,800,000 Years) |
 |
2048^10 |
| 11 | 59.5 Sextillion Years (59,505,237,711,523,000,000,000 Years) |
 |
2048^11 |
| 12 | 1.218 Septillion Years (1,218,667,203,867,850,000,000,000,000 Years) |
 |
2048^12 |
# SEED PHRASE
0.002
seconds
seconds
KEY FINDINGS #2
Click to Tweet
What if AI has 12 words from your seed phrase, but not in the right order?
It only takes AI about
8 minutes to discover
the right sequence of
your seed phrase.
KEY FINDINGS #3
Click to Tweet
The time needed to recover 8 missing words from a seed phrase is 174 times longer than the current age of the universe.
What is LSTM?
LSTM, or Long Short-Term Memory network, is a type of Recurrent Neural Network (RNN) designed specifically to remember information over long sequences, overcoming the limitations of traditional RNNs that struggle with "forgetting" information over time. It’s widely used in applications involving sequential data, such as text, speech, and time-series predictions.
How Does LSTM Work
in Seed Phrase Recovery?
In the context of seed phrase recovery, LSTMs are used to predict the next likely words in a sequence, or even complete missing words, based on patterns learned from previous examples of seed phrases. Here’s a simple way to understand how LSTMs work:
Memory Cells and Gates:
Memory Cells
LSTMs have "memory cells" that store information over time. These cells help the network remember ssential details and discard irrelevant ones.
Gates
LSTMs have 3 types of "gates" that control the flow of information:
- Forget Gate: Decides which parts of the previous words' information are irrelevant to the current context and can be "forgotten" or ignored.
- Input Gate: Determines what new information (next possible words) should be stored based on the sequence seen so far.
- Output Gate: Controls the next word suggestion or prediction
As data flows through an LSTM (for example, words in a sentence), these gates open and close to ensure the right information is stored, updated, or discarded. This allows the LSTM to understand sequences and make predictions based on what it has learned over time.
Key Functionalities of LSTMs for Seed Phrase Recovery:
Learning Word
Dependencies
LSTMs analyze the relationship between each word and its neighboring words. This is crucial for seed phrases, as each word is selected from a fixed vocabulary (e.g., the BIP-39 word list), and their order is essential for recovery.
Sequential
Prediction
During training, the LSTM learns to predict the likelihood of each word following a given sequence. By training on known seed phrases, it can generate or suggest probable words that follow in a partially complete or corrupted seed phrase.
Generating
Missing Words
Given a partial seed phrase, the LSTM can fill in missing words by understanding common patterns in seed phrase sequences. For example, if a user remembers only a few words of their seed phrase, the LSTM can suggest words that typically appear before or after known words in the sequence.
How Can You Protect Your Seed Phrase?
Never Share Your Seed Phrase
Your seed phrase is the key to your crypto wallet, granting full access to your funds. Never share it with anyone, even with individuals or platforms claiming to offer support or security services. Reputable services will never ask for it, and sharing it puts your assets at risk. Treat your seed phrase like the password to a secure vault—it should stay private and confidential at all times.
Write it Down and Store in a Secure Location
Store your seed phrase by writing it on paper or engraving it on a durable material designed to withstand damage, and place it in a secure, secret location. Physical storage protects it from online threats, but ensure that only you (or trusted individuals) know where it’s stored to prevent unauthorized access.
Use Redundant and Distributed Backups
Consider having multiple physical copies of your seed phrase stored in different secure locations. This redundancy protects against accidents like fire, flooding, or theft at one location. Distributing copies across multiple safe places also ensures you can access your funds in an emergency without risking complete loss.
Methodology
For this campaign, we use 2048 seed phrase words from the BIP39 word list to yield 85,714,285 combinations of seed phrases for analysis. Using a cloud-based server, we trained a neural network for 30 days to predict the correct seed phrase and estimate the recovery time. To ensure accuracy, we split the combinations into 80% for training and 20% for testing, then further analyzed the recovery time by examining scenarios from one to twelve missing words. The predictive model leverages Long Short-Term Memory (LSTM) architecture, capable of achieving an average prediction rate of up to 994,051 guesses per second.