A thief who takes your seed plate leaves an empty space, and you move the coins that night. Someone who photographs it and puts it back leaves nothing. Tamper evidence is the work of turning that second case into the first: making a copy leave a mark you will see.
It does not stop a copy. It tells you one may have been made, early enough to move the funds first.
What a seal proves
Roger Johnston's Vulnerability Assessment Team at Argonne National Laboratory separates the two: a lock delays entry, and a seal "leaves behind evidence of unauthorized entry." Pulling a seal off is not defeating it. Defeating a seal means opening it and closing it again, with the original or a counterfeit, without anyone noticing.
That is easier than the packaging suggests. The Argonne team studied 244 seal designs, from cheap adhesive labels to electronic seals, with prices spread over a factor of 10,000. One person using low-tech methods defeated the median seal in 43 seconds, with $5 of tools and supplies. Price didn't predict much: an earlier Los Alamos study by the same group found that the cost of a seal "did not correlate well with its effectiveness." DEF CON's Tamper-Evident Village has attendees practise on adhesive seals, envelopes and evidence bags, to show how many "can be tampered with without leaving evidence."
Johnston's summary of why: defeating seals "is mostly about fooling people, not defeating hardware." In a 2006 packaging test, 71 tamper-detection experts tried to spot the tampered items. Their results were statistically the same as random guessing.
So a seal buys you something narrower than it looks. It forces an attacker to have skill, time and a plan. The brother-in-law with ten minutes and a phone camera usually has none of those. A professional who knows what you own and has a week might. Set your expectations by who you are actually defending against.
The same Argonne work has the good news: most seal attacks have cheap countermeasures, as long as the person inspecting knows what to look for. Here, that person is you.
Hiding is not encrypting. A seal tells you a backup was opened. It does nothing about what was read inside. Put a BIP-39 passphrase on the wallet and keep it somewhere other than the sealed backup, so a seed copied from the bag is not enough on its own.
Numbered bags and seals
Tamper-evident bags are the kind banks, couriers and evidence rooms use: a printed serial number and a closure built to show damage when opened. Johnston lists two minimum requirements for any seal: it needs a unique identifier, and somebody has to inspect it.
To seal a backup:
- Wrap the plate or card in an opaque envelope first, so the words can't be read or photographed through a clear bag.
- Put the envelope in the bag and seal it as the maker's instructions say. Sign across the closure with a permanent pen.
- Write the bag's serial number, the date, and what is inside ("plate, copy 2 of 2") in a log kept away from the bag.
- Photograph the sealed bag, front and back, with the serial number and the closure in sharp focus. The seed itself never goes in front of a camera; see camera, screen and clipboard leaks for why.
- Store the photo somewhere the person you worry about can't reach: a second device, a printout at a relative's, a box at the bank. A photo kept next to the bag can be replaced along with it.
Buy more bags than you need, from a supplier who doesn't know why, and keep the spares out of the house. Spare bags in the same drawer are a counterfeit kit for an insider. Johnston lists "put on a different kind of seal with the correct original serial number" as a standard attack, and the photo is how you catch it.
Glitter and random patterns
A printed serial can be copied. A random pattern is much harder. The DIY version is glitter nail polish: paint it over a seam or a screw, let it dry, photograph it close up, and compare later. Trammell Hudson, who credits Eric Michaud and Ryan Lackey with suggesting it at the 30C3 congress in December 2013 and used it on his laptops' screws, describes "blink tests": flip between the old photo and a new one, and any change in the pattern jumps out.
For a seed backup:
- Seal the envelope flap or the bag's closure, then paint a stripe of glitter polish across the join so it covers both sides.
- Let it dry for a day. Photograph it close up, square on, in even light, with a coin or ruler in frame for scale.
- Store the photos away from the backup, as above.
- At each check, take a new photo from the same angle and flip between the two. The glitter should match flake for flake.
Glitter makes a counterfeit hard to build. Check the edges of the patch as well as the pattern.
A fingerprint for the seed
Seals catch a backup that was opened. They don't catch a quieter attack: someone takes your backup and leaves one holding a different seed, in a counterfeit seal good enough to pass. Every seal check comes back fine, and you find out the day you restore and see an empty wallet.
The check for that is the wallet's master key fingerprint. BIP-32 defines it as the first 32 bits of the hash of the master public key: four bytes, usually shown as eight hexadecimal characters. Some hardware wallets show it on screen; check whether yours does. It changes completely if any word or the passphrase changes, and it reveals nothing that lets anyone spend.
- When you set the wallet up, write down the fingerprint the hardware wallet shows on its own screen. Keep it in the log.
- When you test the backup, restore it on a wiped device and compare the fingerprint it shows with the one you wrote down.
- If they differ, the backup is wrong: miscopied, or replaced.
Four bytes is about 4.3 billion possibilities. That is plenty to catch a mistake or a casual swap. It is not a cryptographic proof against a well-funded attacker, so treat a matching fingerprint as one check among several.
Never compute a hash of the seed words yourself on a phone or computer. Typing the words into a hashing tool puts them on a connected device, which is the leak you are trying to detect. The fingerprint is useful because the hardware wallet computes it, and the words stay on the device.
A check schedule
A seal nobody inspects proves nothing, and inspecting has a cost: every visit to the hiding place shows someone where it is. Pick a rhythm and stick to it.
- Every three to six months, on a date you set, check each sealed copy alone, with the reference photos to hand.
- After any event that gave someone access: a long stay by a guest, a contractor, a move, a break-in, a breakup.
- Each time, compare serial, closure, signature and glitter against the photos. Write the date and "intact" in the log.
- Once a year, or when you test the backup anyway, compare the fingerprint.
A copy kept out of the house, in a bank box or with a relative, gets checked when you are there for another reason.
When a seal is broken
Assume the seed has been seen. It may have been a cat or your own clumsy hands. You can't tell, and guessing wrong costs everything in the wallet.
- Don't open the bag yet. Photograph it as you found it.
- Make a new wallet on a wiped device with a new seed. Write the new backup and test it before anything moves.
- Send all funds from the old wallet to the new one, today.
- Then open the old bag, see what you can learn, and destroy the old backup.
- Seal the new backup with fresh bags and new photos, and think about who had access. If it was someone close, the new copy goes somewhere they have never been.
A seal that might have been defeated, or a fingerprint that doesn't match, gets the same response. Moving funds costs a transaction fee and an afternoon, which is nothing next to what a copied seed can cost. See seed phrase storage and split a seed across places for where the new copies go.
We will never ask for a seed word or a passphrase. Never type one into this site or any tool it links to.
Sources
- Roger G. Johnston and Jon S. Warner, Better Approaches to Physical Tamper Detection, Vulnerability Assessment Team, Argonne National Laboratory, ShmooCon 2010: definitions of a seal and of defeating one, the 244-seal results (median 43 seconds, $5 of tools), the attack list, the 2006 packaging test with 71 experts, and the countermeasures note.
- Roger G. Johnston, Anthony R. E. Garcia and Adam N. Pacheco, Efficacy of Tamper Indicating Devices, Los Alamos National Laboratory, April 2002 (NCJ 194397): seal cost did not correlate well with effectiveness.
- DC24 Tamper-Evident Village, DEF CON forums: the village's stated goal and its work on adhesive seals, envelopes and evidence bags.
- Eric Michaud and Ryan Lackey, Thwarting Evil Maid Attacks, 30C3, 30 December 2013.
- Trammell Hudson, Glitter: nail polish on laptop screws, close-up photos and blink tests.
- BIP-32: Hierarchical Deterministic Wallets, "Key identifiers": the key fingerprint is the first 32 bits of the identifier.
- BIP-39: Mnemonic code for generating deterministic keys, the passphrase.