Wallets
How hardware wallets work, and what they cannot stop
A hardware wallet is a small single-purpose computer that holds private keys and uses them to sign without letting them out. That puts the keys out of reach of software running on the computer it is plugged into. Three ways of losing funds pass straight through it: approving a transaction you cannot read, using a device that was altered before it reached you, and the recovery phrase, which is a complete copy of the keys kept somewhere else.
What the device does
Why moving the signing step off a computer matters is covered in hot wallet, cold wallet. This piece is about the device.
Set-up happens on it. The device generates the secret from its own randomness, shows the recovery phrase on its own screen for you to write down, and keeps the keys. A genuine device never arrives with that work done. Ledger, warning about second-hand units in 2018, put it plainly: its documentation “does not include any pre-existing seed words or pin code”, and a device that comes with them “may have already been used by somebody else” (Ledger).
Use follows a fixed loop. The wallet application on a computer or phone builds an unsigned transaction and passes it to the device. The device shows what it has been asked to sign, you confirm on the device itself, and it hands back a signature. The key stays on the device throughout, so malware on the computer has nothing to copy. The first-hour routine in your first crypto wallet, a restore test, a small send and a send back, applies to a hardware wallet unchanged.
Two consequences are easy to miss. Losing the device does not lose the funds, because the phrase restores the keys on any compatible device; equally, anyone holding the phrase has the funds without touching the device. And every device runs firmware that its maker writes, signs and updates. Ledger describes a “genuine-check mechanism” with which users “can verify they are running the legitimate operating system and embedded applications” (Ledger Academy). A check of that kind confirms the software is the maker’s. Using the device means trusting the maker as well as the hardware.
The screen is the security feature
The device’s screen is the one display in the chain that the computer cannot draw on. Ledger says its secure element “is responsible for driving your Ledger device’s secure screen”; Trezor says its trusted display “shows the real data your Trezor is working with, on hardware that malware can’t reach” (Trezor).
Trezor’s guide also states the limit. The display “confirms what your device is doing: the real address it generated, or the real transaction it’s about to sign. It can’t tell you whether an address someone gave you is the right one.” Checking a receiving address on the device proves the address belongs to your wallet. Checking a destination on the device proves nothing swapped it after you pasted it. Neither proves you copied the right address in the first place, which is the gap address poisoning uses.
Contract calls are the larger gap. For a plain transfer the device can show an amount and an address. For a call to a smart contract it can show only what it knows how to decode. Ledger defines blind signing as “the act of approving a digital transaction without being able to verify its full details in a human-readable format”, where the screen may show “a complex cryptographic hash or a generic message like ‘Data Present’”. At that point the device is still protecting the key and has stopped protecting the decision.
Clear signing, where the device decodes a call into its function, amounts and recipients, depends on someone describing each contract to it. ERC-7730, still a draft, proposes a shared JSON format for those descriptions. Where the software has no description, Ledger’s glossary says, a hardware device “may simply show ‘Data Present’”, and the rest is left to trust. Bybit’s signers, on NCC Group’s reading of the February 2025 theft, “blind-signed the messages without carefully checking their contents”; multisig wallets explained covers that case.
The secure-element question
A secure element is, in Trezor’s description, “a tamper-resistant chip designed to securely store secrets and enforce access controls”, “the same type of chip used in credit cards and passports” (Trezor). Its job is the case that malware on your computer cannot reach anyway: somebody holding the device.
Manufacturers use it differently. Ledger says its secure element “generates and stores your private keys”, drives the screen and runs the device’s operating system. Trezor’s Safe 3 and Safe 5 keep the keys encrypted on the main chip. Their secure element holds a secret the encryption depends on and “only releases the secret if the correct PIN is entered”; after 16 wrong PINs it erases the secret and the device resets. Trezor’s own summary is that all its devices protect keys from online attacks, while those two models “add extra protection against physical attacks with a Secure Element”.
Certification is the usual evidence offered. Ledger cites Common Criteria EAL5+ for its Nano X and EAL6+ for the Nano S Plus and Stax; Trezor cites EAL6+ for the element in its Safe 3 and Safe 5. One difference Trezor draws is who else can inspect the chip: it notes that “many Secure Elements are locked behind non-disclosure agreements that prevent public security research”, and gives that as its reason for choosing one with public documentation.
None of this touches what gets signed. A secure element answers the question of what happens if someone steals the device and has time with it, which matters to anyone who carries one. It has no bearing on any incident in the next two sections.
Where the device came from
Ledger’s statement of 15 October 2026, responding to a hardware implant attack, describes what happens when the device itself is altered before it reaches the buyer. A criminal group, “over many months”, bought genuine Ledger devices, opened them, “fitted custom-built hidden components designed to capture the recovery phrase during setup, and put them back into distribution through resellers in South East Asia”, then drained wallets in one coordinated operation. Ledger says there was “no flaw in Ledger’s chip, operating system or servers”, and that this “was not a systemic supply chain attack, it was a localized distribution attack”; units its teams bought from stores in other markets, Germany included, and opened were clean, as were samples from warehouses. It asked authorised resellers and distributors to suspend sales in Southeast Asia and Hong Kong, reminded them that returned devices “can never be resold in any form”, and told them to keep stock off shelves and displays. The statement gives no loss figure and no count of victims, and says fuller details will be published later. DefiLlama’s hacks record, as of 22 November 2026, lists “Ledger (Supply Chain)” at $93.0M, dated 9 October 2026, under Frontend & Infrastructure.
An authenticity check confirms that a device’s chip and software are the maker’s. By Ledger’s account the altered units were genuine devices with components added, and its statement does not say whether any check could have revealed them.
The low-technology version is older. In Ledger’s 2018 warning, a buyer received a second-hand device that the seller had already initialised, together with the 24 words, so the coins on it were accessible to “whoever had initialized this device and possessed a copy of the 24 words”. Nothing about the hardware had to be defeated.
Both cases sit before the moment the phrase is generated, which is where a hardware wallet’s protection begins. What happened to the unit before then, who handled it and whether it was returned, displayed or opened, is outside anything the device can vouch for. A phrase that arrives printed, on a card or in a letter, rather than appearing on the device’s screen during set-up, means someone else has it; Ledger’s position is that such a device should not be used as it is.
What the vendors’ disclosures show
Two of Ledger’s own disclosures show where the risk moves when the device holds. The first was a customer-data leak. In July 2020 Ledger disclosed that “an unauthorized third party had access to a portion of our e-commerce and marketing database through an API Key”. The exposure was “approximately 1M” email addresses and, for 9,500 customers, also names, postal addresses, phone numbers or products ordered (Ledger, 29 July 2020). In December the database was posted on a forum, and Ledger’s chief executive wrote that it held detailed information on “approximately 272,000” customers, more than the company’s own logs had shown (Ledger, 21 December 2020).
There were no keys or phrases in it; Ledger said the breach concerned contact and order details only. There was a list of people who owned hardware wallets and, for about 272,000 of them, where they lived. The chief executive’s message acknowledged that “many of you have been targeted by e-mail and SMS phishing campaigns”, and Ledger’s page on phishing campaigns describes physical letters that “prompt individuals to scan a QR code and/or go to the website mentioned in the letter”, ending in a request for the 24 words. Neither page says where the senders got the addresses.
The second reached users through software. On 14 December 2023, according to Ledger’s incident report, a former employee “fell victim to a sophisticated phishing attack” that gave the attacker their npm account through a session token, bypassing two-factor authentication. The attacker published malicious versions 1.1.5, 1.1.6 and 1.1.7 of Ledger Connect Kit, a JavaScript library that decentralised applications load at runtime to talk to Ledger devices. The code carried a drainer, Angel Drainer, which Ledger describes as “specifically designed to craft malicious transactions that are draining signers, also known as hardware wallets, when signed”. A genuine version was deployed within 40 minutes of Ledger becoming aware; the malicious file stayed reachable for about five hours through caching, and Ledger estimates assets were actively drained for “less than two hours in total”. DefiLlama’s record lists the incident at $484K, in the same class.
The devices kept the keys and signed what they were asked to sign. Ledger’s remedy, in its words, was to “double down on preventing Blind Signing” and to push clear signing. Wallet drainers explained covers what a kit like that asks for.
What it leaves to the owner
A hardware wallet answers one question, whether software on your computer can take your keys, and hands three back: what you are approving, where the device came from, and where the phrase is. The phrase is the subject of can a seed phrase be hacked and where to store a seed phrase, and a passphrase adds a second secret on top of it. Compromised software around a wallet, fake apps and altered extensions, is one of the routes in how crypto wallets actually fail.