A seed backup has to pass two opposite tests.
You—or an authorised heir—must still be able to find and use it years from now. A thief, compromised phone, house guest or cloud account must not be able to find and use it today.
Most seed-storage advice optimises only one side. “Hide it well” can create an unrecoverable wallet. “Make several copies” can turn one secret into several theft opportunities. “Use metal” protects against fire but does nothing about discovery.
The design problem is to survive destruction, discovery and forgetting at the same time.
Start with the three failure domains
| Failure | What it looks like | Control that actually addresses it |
|---|---|---|
| Physical destruction | Fire, water, corrosion, accidental disposal | Durable material or a second location outside the same physical event |
| Unauthorised discovery | The complete seed is found, copied or photographed | Access control, separation, threshold design where justified |
| Recovery failure | Words are wrong, passphrase is forgotten, heirs cannot reconstruct the process | Verified backup, documented recovery and periodic rehearsal |
A storage decision should name which failure it solves. If it cannot, it is probably adding complexity rather than resilience.
One verified backup is the starting point
Record the recovery words in the exact order generated by the wallet and verify them before the wallet holds a meaningful balance. Use the manufacturer’s recovery-check function or a controlled restore procedure appropriate to the device.
The important word is verified. A beautifully engraved backup containing one wrong word is durable evidence of a mistake.
Keep the recovery material offline. Do not photograph it, email it, store it in ordinary cloud notes, paste it into support chat or type it into a website. A normal self-custody seed is bearer authority: someone who obtains the complete secret, plus any required passphrase, may be able to recreate the wallet elsewhere.
Paper and metal solve a material problem, not an access-control problem
Paper can be perfectly reasonable in a controlled environment. It is cheap, easy to inspect and easy to replace before meaningful funds arrive. Its weakness is physical durability.
Metal improves resistance to heat, water and crushing when the recording method and product design actually survive those conditions. It does not make the words less powerful.
A metal seed found by the wrong person is simply a more durable stolen secret.
Choose the material after identifying the physical threat. Do not confuse “fire resistant” with “secure.”
Add a second copy only if it escapes the first copy’s failure
Two backups in the same drawer are two copies and one failure domain.
A second copy earns its place when it changes the failure map. For example:
- a home backup plus a controlled off-site location can survive one building loss;
- paper plus metal can reduce common material failure;
- a documented threshold backup can prevent one location from holding the full secret;
- a tested replacement signer can reduce dependence on one device without creating another seed copy.
Every extra full seed also creates another discovery surface. Redundancy is useful only when the new failure it prevents is worth the new exposure it creates.
Do not invent your own word-splitting scheme
Cutting a 12- or 24-word phrase into homemade fragments feels clever because no one paper contains the whole list. It creates problems that are difficult to see until recovery day: unclear thresholds, lost ordering, partial brute-force exposure, no standard software path and instructions that heirs may not understand.
If no single location should hold the complete recovery secret, use a documented threshold design such as SLIP-39 where the wallet supports it, or consider multisignature when the goal is to distribute spending authority rather than merely backup material.
Those are not interchangeable:
- Multi-share backup distributes information required to reconstruct one wallet secret.
- Multisig distributes independent signing keys so one key may be insufficient to spend.
Use the simpler architecture that solves the threat you actually have.
A passphrase creates a second recovery problem
A BIP-39 passphrase can derive a different wallet from the same recovery words. That can reduce the consequence of seed discovery when the passphrase remains independent and strong.
It can also produce a perfectly valid empty wallet when entered incorrectly. That makes forgotten spelling, spacing or inheritance instructions a permanent failure mode.
Do not store the passphrase beside the seed. More importantly, do not add one until you can explain how it will be recovered years later without giving the same person or location both secrets today.
Design the location around who can enter it
A home safe, bank safe-deposit box, trusted off-site property and professional custody arrangement fail differently.
A home safe is immediately accessible but shares household and regional risks. A bank box separates the backup from the home but adds institutional access rules and availability constraints. A trusted property adds another person’s behaviour to the system. A professional or collaborative custody service may distribute recovery while adding contractual and identity dependencies.
There is no universally correct hiding place. Write down the people and events that can reach each location. That is more informative than rating a container “secure.”
Your future self is part of the threat model
A recovery system can be theft-resistant and still fail because nobody remembers what it is.
Without writing the seed itself into ordinary records, document enough context to answer:
- which wallet or backup standard is involved;
- whether a passphrase or threshold is required;
- where authorised recovery instructions can be found;
- which device or compatible software can restore it;
- who should be involved if the owner is unavailable;
- what must never be typed into an ordinary website or support form.
An inheritance plan should reveal the path at the right time without revealing the complete secret too early.
The recovery drill is the acceptance test
Before treating the setup as finished, prove three things:
- Readable: the recorded words or shares are complete and unambiguous.
- Recoverable: the documented process recreates the expected wallet or verified account fingerprint.
- Separated: loss of one location, one device or one person does not accidentally expose or destroy every recovery route.
Repeat the review after moving house, changing banks, changing family arrangements, replacing hardware or materially increasing the balance.
If the seed is ever exposed, storage has stopped being the problem
Once another person may possess the complete seed, the correct response is not a better safe. The wallet authority itself has to change.
Create a new wallet with new recovery material and migrate the assets. The incident process is covered in Seed Phrase Compromised?.
For material selection, see the metal seed backup guide. For succession design, use the crypto inheritance plan.
Primary sources
- BIP-39 mnemonic specification
- SLIP-39 Shamir backup specification
- Jameson Lopp metal backup tests
- Trezor wallet backup documentation
No affiliate link is used in this article.
