What is a hardware wallet? How the device protects your keys

A hardware wallet is a small device that generates the private key to your cryptocurrencies, stores it and never releases it. It looks like a USB stick or a chip card, usually carries a small display, and is connected to a computer or a phone only when you want to sign something. Everything that follows explains why this particular design makes the difference.

Which wallet types exist, how they differ and which one suits your amount is covered in our guide How to store crypto: crypto wallet types compared.

The question has a practical side right now. Reports of tampered devices, fake manufacturer websites and drained wallets have piled up through October 2026. To work out which of those attacks a device can actually fend off, you need to know what happens inside it.

A hardware wallet holds keys, not coins

The first and most important point contradicts the name. There is no money inside a wallet. Bitcoin and every other cryptocurrency exist purely as entries in a blockchain, a ledger kept on thousands of machines at once. Those entries cannot be moved and cannot be slotted into a device.

What a wallet holds is the private key: a very large number that proves you are entitled to dispose of a particular entry. Whoever holds that key can spend the associated coins. Whoever loses it can no longer reach them, and nobody can reset it. There is no bank here to issue a new password for a forgotten one.

That sets the entire engineering task for such a device: the key has to be usable without ever becoming visible.

The private key never leaves the device

Hardware wallets solve this by generating the key inside the device and keeping it there. Manufacturer Trezor describes the procedure in its own documentation: the private key never leaves the hardware wallet and at no point comes into direct contact with the internet-connected device.

A computer the wallet is plugged into therefore never sees the key, even with malware running on it. It receives only the result: a finished signature. Historically this answers a concrete problem. In the early years, private keys sat in encrypted files on ordinary computers, and malware emptying an entire wallet was nothing unusual. The first hardware wallet, the Trezor Model One, was built in 2014 at SatoshiLabs in Prague as a counter-design.

What really happens when a transaction is signed

The sequence has four steps, and it explains why an infected computer is less dangerous than it sounds.

First you assemble the transfer in an application on the computer or phone: recipient address, amount, fee. Those details go to the hardware wallet. Its display shows them there, independently of the computer. You confirm with a button on the device. Only then does the chip compute the signature and hand it back, at which point the application broadcasts the signed transfer to the network.

The third step is the decisive one. Because the display belongs to the device and not to the computer, malware can propose a false recipient address, but it cannot stop the real address from appearing on that small screen. Read it there and compare, and the swap shows up. Press the button blind, and it does not. The device moves security to a place the attacker cannot reach, and asks for a glance at the display in return.

Secure element and PIN: the two locks inside the device

Two layers of protection work against direct access to the device, and they work differently.

The first is the PIN. Without it the device cannot be operated, and after a manufacturer-dependent number of failed attempts it wipes its memory. A stolen device is therefore worthless to the thief at first.

The second is the secure element, a dedicated chip of the kind also found in bank cards and passports. It is designed to withhold secrets even when somebody grinds it open, measures its power draw or knocks it off its clock with targeted voltage spikes. Not every device on the market carries one; some manufacturers rely on open-source components instead and argue for verifiability over tamper resistance. Which route is better has been disputed in the industry for years, and the answer depends on the attack you want protection from.

Both locks share one limit: they protect the device, and the recovery phrase sits outside their reach.

The recovery phrase: 12 or 24 words under BIP-39

During initial setup the device shows a sequence of words, usually twelve or twenty-four. This recovery phrase, often called a seed phrase, is the key in human-readable form. From it the entire set of keys can be rebuilt at any time, on any device, including one from a different manufacturer.

Behind the procedure stands an open standard called BIP-39, published in 2013 and in use ever since. It defines how a random number becomes a sequence of words and how the words become a key again. That is exactly why no manufacturer can lock you in: if a device breaks, you restore your holdings onto a new one with the same phrase.

The standard is also clear about what it is not meant for. It exists to make randomness generated by the device readable, and expressly not to turn a self-invented sentence into a key. Inventing a phrase instead of having one generated defeats the whole protection.

2048 words, 11 bits per word: where the phrase gets its strength

A short calculation pays off here, because it makes clear why guessing is out of the question.

The standard's word list contains exactly 2,048 entries. 2,048 is the eleventh power of two, so each word carries precisely 11 bits of information. Twelve words give 132 bits, of which 128 are the actual random number and four form a checksum; twenty-four words give 264 bits, 256 of them random and eight a checksum. That checksum is why a device usually spots a mistyped phrase immediately.

128 bits of randomness mean around 340 sextillion possible phrases, a number with 39 digits. At 256 bits it has 78 digits. For comparison, the visible universe holds somewhere around 10 to the power of 80 atoms on common estimates. Working through a 24-word phrase exceeds any conceivable computing power; the limit here is set by physics.

The practical consequence is an uncomfortable one. Since nobody will guess the phrase, every successful attack runs through the human being: a photographed phrase, a fake website, a tampered device. That is exactly where the losses happen.

Hot wallet, software wallet, hardware wallet: where the difference lies

Three terms get mixed up regularly, and the difference always comes down to one question: where does the private key sit?

With an exchange wallet it sits with the provider. You have an account and a password, but no key. If the exchange fails or is attacked, you are a creditor. In exchange you can buy and sell there, and a forgotten password can be reset. Provider terms are set out in our comparison of crypto exchanges.

With a software wallet, or hot wallet, the key sits on your computer or phone, encrypted, but on a device with an internet connection. You control it yourself, while sharing the device with everything else running on it.

With a hardware wallet the key sits on a device that does nothing else and stays disconnected most of the time. The technical terms for this are cold storage, for custody without a network connection, and air gap, for the space in between.

In practice this does not make one of them right and the others wrong. A split is common: the amount being traded stays at the exchange or in the software wallet, and the holding meant to sit still goes into the device. Everyone decides for themselves at what sum the effort pays off; the devices cost between roughly 50 and 200 euros depending on the model.

Blind signing: the gap the device does not close

One weakness concerns the very step that is supposed to carry the security. For simple transfers the display shows the address and amount in plain text. For more complicated operations, such as approving a program on Ethereum, the device runs out of space or of the ability to translate, and shows nothing but an unreadable block of data. Confirm then, and you sign something you have not read. This is called blind signing.

Several manufacturers have improved on it in recent months and can now display considerably more operations in plain text. What can be changed in your own device settings we set out in detail in September.

Supply chain and fake sites: the real attack routes of 2026

This year's attacks target the device's route to the buyer and the person in front of it, leaving the mathematics untouched.

In October 2026 Ledger confirmed that an additional component had been found inside devices where it does not belong; reports of it reached Europe too. We gathered the case and the checks to run on the device. Shortly before that, according to those affected, 86 million dollars disappeared from wallets, and that case too ran through tampered devices rather than broken encryption. In parallel a fake manufacturer site topped the Google results and asked visitors there for recovery phrases.

An earlier case pointed to the same lesson, one in which a faulty random number generator made keys computable and cost 130 million dollars: a device is only as good as its provenance. Buy from the manufacturer or a named reseller, never second hand, and set it up yourself. A device that arrives with a recovery phrase already printed on it is always an attempted fraud; no manufacturer mails out phrases.

What a hardware wallet does not do

Four things are beyond even the best device, and they belong in a complete answer to the question in the title.

It offers no protection against price falls, because it stores value without judging it. It offers none against a transfer to the wrong address, because a confirmed transaction cannot be recalled. It offers none against a phrase somebody else has seen, whether in cloud storage, in a photo or on a note in a drawer. And it offers none against your own forgetting: without the phrase and without the PIN, the holding is gone once the device fails. What to do in that case, and how the tax office treats such a loss, we wrote up separately.

Tax and the holding period: what the move to your own device triggers

For investors in Germany, moving from an exchange to a device of your own is uncritical for tax purposes. A transfer between your own wallets is not a sale and triggers no taxable event; the one-year holding period under Section 23 of the German Income Tax Act keeps running and does not restart.

Two things still need attention. The acquisition date has to remain provable, because gains are tax-free after a year has passed, and the proof is yours to supply. Keep the exchange statements that show date and quantity. Second, allocating holdings remains your job once they sit spread across several wallets.

Our assessment: the device is the smaller half of security

From the newsroom's perspective, a hardware wallet makes sense for anyone holding a position for the longer term and not moving it constantly. The reason differs from the one in the advertising.

This year's evidence supports it. None of the large losses traced back to broken encryption. The causes were tampered devices, a fake site in search results and a faulty random number generator. The protection the device provides mathematically failed in none of those cases.

Nothing argues against the device itself, but plenty against the expectations placed on it. Buy a hardware wallet and then photograph the phrase, and you have spent money for nothing. Security emerges only from the combination: a device from a known source, set up by you, a phrase kept only on paper or metal, and the display read at every confirmation. Drop one of those four and the rest will not carry.

No recommendation for a particular model follows from this, and a total loss remains possible with any cryptocurrency, however well it is stored.

Hardware wallet: the phrase decides, not the device

If you want to take the step, three steps in this order are enough:

1. Choose a device and buy it directly. From the manufacturer or a reseller it names, never second hand and never from a marketplace third party. Which models offer which protection is set out in our comparison of hardware wallets.
2. Set it up yourself and write the phrase down by hand. Paper or a metal plate, two locations, no photo, no cloud storage and no typing it into a computer. For smaller amounts that stay in motion, a software wallet still does the job alongside.
3. Transfer a small amount first. Send it back once before the rest follows, and keep the purchase statements for the holding period. The matching tax tools take the allocation off your hands later.

How randomness becomes a sequence of words and then a key again is set out verbatim in the BIP-39 specification; the description of the signing process comes from Trezor's documentation.

(As of October 14, 2026. This article is not investment advice. Prices and fee structures change; check the terms with the provider before you buy.)