Ledger vs Trezor vs Tangem: The Seed Phrase Is the Leak

Two Implants, One Reseller, and the End of the Chip Debate

On October 30, 2026, Trezor told its customers something that should change how everyone thinks about hardware wallets. A Trezor Safe 5, bought in August from the Southeast Asian reseller CryptoBilis through the Indonesian marketplace Tokopedia, had turned up with an extra circuit board wedged between its main board and its screen. The board carried a cellular module and an antenna. According to the customer report Trezor received two days earlier, it reads the screen, including the recovery phrase that Trezor calls the wallet backup, and sends it out over a mobile network every time the device is plugged in. Trezor said the implant “appears to work like the implants seen in Ledger devices,” though it would not draw a direct comparison mid-investigation “because each maker builds its devices differently,” per The Hacker News.

That phrase, “the implants seen in Ledger devices,” is the whole story. Three weeks earlier the same reseller had shipped tampered Ledger units whose hidden boards did the same job: a microcontroller, an eSIM, and an LTE modem that captured the 24 words as they appeared on screen during setup and texted them to an attacker. Ledger confirmed the implant “recognized and decoded” on-screen recovery phrases and “transmitted them over LTE,” while leaving the secure element, the firmware, and every signature untouched. On-chain trackers put the losses somewhere between roughly $86 million and $93 million across more than 300 wallets before Tether froze $10 million of USDT, CoinDesk reported.

The detail that matters for anyone choosing a wallet in 2026 is that the attack did not care which brand it was inside. It did not break Ledger’s certified, closed secure element, and it did not need to break Trezor’s open one either. It sat beside the chip and read the one thing both devices are built to display: your seed phrase.

Bitcoin traded around $77,381 on the morning of the Trezor warning, down about $2,655 from the day before and roughly $31,000 below where it sat a year earlier, according to Fortune. Prices move; the lesson does not. For most of a decade the hardware-wallet argument has been a chip argument: whose secure element is rated higher, whose firmware is more auditable, whose design is harder to crack in a lab. The implants sold through CryptoBilis answered that argument by going around it. What follows is a comparison of Ledger, Trezor, and Tangem built around the question those implants actually posed: how do you keep a secret that has to be displayed to be backed up, when the thing displaying it might be compromised?

Three Answers to One Question

Every self-custody wallet has to solve the same problem. Your private keys must never touch an internet-connected computer, yet you still need a way to recover them if the device is lost, and you need to see and approve what you are signing. The three best-known brands in 2026 solve it in three incompatible ways, and the CryptoBilis implants put each design under a spotlight.

Ledger’s answer is a sealed black box. Its keys live inside a bank-grade, certified secure element, and its operating system, BOLOS, is closed-source. You cannot read the code, so you trust Ledger and the certification labs instead. Late in 2025 the company even stopped calling its products “hardware wallets” and started calling them “signers,” arguing that coins never live on the device; the device only signs.

Trezor’s answer is the opposite: show everyone everything. Its firmware is fully open-source, and its newest flagship, the Safe 7, is built around TROPIC01, which the company bills as the first openly auditable secure element with no non-disclosure agreement attached. The logic is that a secret kept by obscurity is not kept at all.

Tangem’s answer removes the secret you have to protect in the first place. A Tangem wallet is a contactless card. Its key is generated on an EAL6+ chip on the card and, by default, never leaves it as a written 24-word phrase. You tap the card to a phone to sign. There is nothing to write on paper, and, as the CryptoBilis episode made painfully clear, nothing for a screen-reading implant to photograph.

Here is how the current lineups and prices compare before the security arguments begin.

The Lineup and What It Costs

The three brands ladder from a sub-$60 entry device to a $399 luxury object, and in almost every case the extra money buys polish, bigger screens, and radios rather than a materially stronger secure element. A Ledger Nano S Plus protects a key with the same class of certified chip as the $399 Stax; a Trezor Safe 3 uses the same security model as the Safe 5. The figures below are manufacturer list prices in late 2026, which shift with bundles and promotions. The Safe 7 launched in Prague in October 2025 at $249, and Ledger’s E Ink Nano Gen5 arrived around the same time at $179.

What the Implant Actually Did

The CryptoBilis implants are worth understanding precisely, because the precise mechanism is what makes them brand-agnostic. A teardown of a Ledger unit, reported by security researchers, found a tiny microcontroller, a small eSIM, and a shielded LTE module, powered by a slightly modified battery, with a wire antenna running along the cell. The board sat behind the display and tapped the data lines feeding the screen. During setup, when the device shows you the 24 recovery words one at a time so you can write them down, the implant read those words straight off the display and transmitted them over the mobile network. The attacker never needed your computer, your phone, or your PIN.

Ledger’s own security lab stressed what the implant did not do: it “does not modify the Secure Element or the Ledger operating system, and does not alter transactions or signatures.” The chip’s attestation, Ledger’s “genuine check,” verifies that the secure element is authentic, but as independent analysts noted, it cannot detect unauthorized physical modifications such as a spying implant if that element itself remains intact. The implant left the chip alone and read around it.

Mark Karpelès, the former Mt. Gox chief executive who now runs a security firm, called the device “quite an evolution” and summed up the asymmetry bluntly to BleepingComputer: “Ledger Nano X are used to store crypto. The spy implants are used to steal whatever crypto you store.” HOGE Wire has tracked this wave of screen-reading hardware and fake-alert phishing across the month in Nobody Cracked a Key: Crypto’s October 2026 Wallet Wave.

The Trezor report, still awaiting lab confirmation at the time of writing, describes the same architecture: a board between the main board and the screen, a cellular module, an antenna, and the same behavior of reading the backup off the display whenever the device is powered. Trezor said it is treating the report as credible while it waits for verification, and urged anyone who bought from CryptoBilis since March 2026 to move their funds now: “If you already use the wallet, move your funds to a new wallet with a new recovery phrase as soon as possible.”

The Seed Phrase Is the Artifact Everyone Agreed to Trust

Step back and the common thread is not the chip, the brand, or even the reseller. It is the seed phrase. Since 2013 the industry standard has been BIP39: a wallet generates a 128- or 256-bit secret, encodes it as 12 or 24 ordinary words, and shows them to you so you can copy them onto paper. That list is the master key to everything. It is portable, human-readable, and vendor-neutral, which is exactly why it became the gold standard of self-custody, and exactly why an implant that can read a screen finds it so useful.

A 24-word phrase has to be displayed at least once to be backed up. That single moment of display is the exposure. Write it on paper and a camera in the room sees it; type it into a “recovery” app and malware sees it; show it on a tampered screen and the implant behind the glass sees it. Every mitigation the seed-based brands offer, whether metal backup plates, passphrases, or Shamir splitting, addresses the paper after the fact. None of them removes the instant the words appear.

This is the argument Tangem has been making for years, and the CryptoBilis incident handed it a live case study. In a note on the episode, Tangem wrote that “no recovery phrase is displayed during setup unless the user explicitly opts into BIP39 mode,” which “removes the specific exposure path the CryptoBilis implant reportedly exploited: there is no phrase to intercept if no phrase is generated.” Its companion app, the company added, “performs a cryptographic authenticity check and warns if a card appears to have been activated before setup, which functions as a supply-chain tamper indicator,” according to Tangem’s learning-hub note.

That is a real structural advantage against this specific attack, and it deserves to be stated plainly. It is not a universal one. Seedless means your key exists only on the cards; lose all of them and there is no phrase to restore from somewhere else, which trades one failure mode (theft of a displayed secret) for another (loss of every physical card). It also means you cannot move your wallet to a competitor’s device, because there is no portable phrase to import. The seed phrase is a liability at setup and an asset at recovery. Which matters more depends on who is holding it.

The Box Is the Attack Surface

If the seed phrase is the target, the delivery van is the reseller. CryptoBilis was not a counterfeit operation selling fakes in a back alley; it was an authorized reseller for both Ledger and Trezor, serving Indonesia, Malaysia, and the Philippines, with devices listed on mainstream marketplaces. The tampering happened somewhere between the factory and the buyer, in the part of the journey no secure element can see. This is the sense in which, as HOGE Wire has argued about a very different exchange breach, a loss can come from neither code nor keys but from the plumbing around them.

Trezor’s response was structural rather than technical. On October 13 it announced it was sharply cutting its authorized-dealer network, stopping new shipments to affected resellers on October 15 and ending their authorization on January 1, 2027. “Following recent supply-chain incidents, we decided to reduce the number of resellers so we can more effectively check those we work with,” the company said. Its chief commercial officer, Danny Sanders, reduced the principle to one line in comments reported by CoinCentral: a hardware wallet is “only as safe as the path it travels to reach its owner.”

Ledger reached the same conclusion from the other direction, asking resellers across Southeast Asia and Hong Kong to suspend sales while it traced the tampering, and chief executive Pascal Gauthier framed the open question as a distribution problem: the “investigation is ongoing to determine who tampered with these signers or where in the distribution chain it happened.” Ledger also stressed that its own systems and direct sales channels were not compromised. The practical advice from all three camps converged: buy directly from the maker, or from a retailer you can verify, and treat a device that arrives already set up, or that asks you to use a “provided” recovery phrase, as compromised.

This is the cleanest takeaway of the whole episode. The single most effective thing a buyer can do costs nothing and favors no brand: control the chain of custody. Order direct. Check the tamper-evidence. Generate a brand-new key yourself, on the device, in private, and never accept one that someone else could have seen.

Sealed: Ledger’s Case for the Certified Black Box

Ledger’s pitch is certification. Every current model, from the $79 Nano S Plus to the $399 Stax, runs its keys through an ST33-series secure element rated to Common Criteria EAL6+, the kind of chip used in passports and payment cards, paired with the closed-source BOLOS operating system. You cannot audit BOLOS, and Ledger argues you should not need to: the chip is certified by independent labs, and keeping the firmware closed denies attackers a map.

Chief executive Pascal Gauthier has leaned hard into a philosophy of dedicated hardware over general-purpose software. “Everything that is in software is close to impossible to protect,” he wrote this year; “you cannot ask a general-purpose phone or laptop to protect your most sensitive secrets.” His bolder claim, that “the safest place to store value is Bitcoin on a Ledger,” sits awkwardly beside his own admission, in the same essay, that across the industry “hackers are winning.”

The sealed model has two well-documented costs. The first is that you are trusting a company, not a codebase, and Ledger’s record on trust is uneven. A 2020 marketing-database breach exposed around a million email addresses and some 270,000 physical mailing addresses, feeding years of phishing and extortion, and in August 2026 a user filed a proposed class action in New York seeking at least $500 million over that breach and the 2023 Connect Kit supply-chain incident. The second cost surfaced with Ledger Recover, the optional paid service that shards an encrypted seed across custodians: technically sound to some, but proof to critics that a closed device can be made to export a key the user thought was captive.

Open: Trezor’s Case for Auditable Everything

Trezor, built by the Czech company SatoshiLabs, made the opposite bet. Its firmware is fully open-source, and chief executive Matej Žák states the thesis without hedging: “We are the most secure because we are the most open-source.” The company’s frustration with the sealed model is specific. “In the past, we used chips under NDA,” Žák told DL News, “and if we found a problem, we couldn’t warn anyone.” The Safe 7’s TROPIC01 element, co-developed with Tropic Square, is pitched as the fix: a secure chip whose design can be inspected without a non-disclosure agreement.

Openness has its own history of costs. Trezor’s older Model One and Model T carried no dedicated secure element at all, which left them exposed to a physical glitching attack that a well-equipped lab could run in about 15 minutes; the Safe line added certified elements precisely to close that gap. And openness did nothing to stop the company’s worst 2026 week off the device: a breach at its third-party fulfillment partner ShipMonk eventually exposed personal data on roughly 81,000 customers, traced to a critical SQL-injection flaw in an analytics tool. Trezor’s repeated line through that episode, that all Trezor devices remained secure, was true and beside the point; the leak was names, emails, and shipping addresses, the raw material for phishing and, now, for the doorstep attacks that follow.

The CryptoBilis implant is the sharper irony. Open firmware let independent researchers discuss the Trezor tampering openly within days, which is the system working as designed. But it also showed that an open codebase is no defense against a board soldered next to the screen. You can audit every line Trezor ships and still lose everything to hardware Trezor never shipped.

Seedless: Tangem’s Case for Never Writing It Down

Tangem is the outlier in form and philosophy. There is no screen, no battery, and no cable: just a card (or a ring) with an NFC antenna and a Samsung S3D232A secure element rated EAL6+. When you set it up, the chip generates its own key and, at your option, clones it to one or two backup cards in the same session. After that there is no seed phrase in the world unless you deliberately turned on BIP39 mode. You sign by tapping the card to your phone.

The design choices cascade from seedless. There is no firmware-update mechanism, which Tangem frames as deliberate: a card that cannot be updated also cannot be tricked into running malicious firmware, though it equally cannot be patched if a flaw is found. Backup is physical redundancy, two or three cards kept in different places, rather than a phrase in a safe. The cards have been audited by Kudelski, Riscure, and, in 2026, Cure53. Co-founder Andrey Kurennykh argues the model is simply honest about human behavior: most people back up a seed phrase badly, and a secret you never create cannot be stolen, lost, or mis-copied.

The trade-offs are real and worth naming. Non-updatable firmware means a discovered flaw cannot be fixed in the field. The absence of a portable phrase means no escape hatch to another vendor and total dependence on the cards you hold. And seedless does not mean coercion-proof: a holder with a card and a PIN can still be forced, at the door, to tap and transfer. What seedless does defeat, cleanly, is the exact attack that drained CryptoBilis buyers: there was never a phrase on a screen for an implant to read. For holders whose nightmare is a leaked or photographed seed, that is the whole argument.

The Lab Attacks Nobody Lost Money To

There is a second kind of attack on these devices, and it is the one the marketing fights over: breaking the chip itself. In 2026 Ledger’s in-house security lab, Donjon, went after both rivals with laser fault injection, firing precisely timed light pulses at a decapped chip to make it skip an instruction. The results made headlines and, tellingly, cost no one a cent.

Against Tangem, disclosed in February and published in July, Donjon used a nanosecond laser pulse to fault a single conditional check in the card’s SetPin routine, resetting the access password without the original password or a backup card and opening the way to drain the card. The catch is the setup: physical possession of the card, a lab rig worth about $250,000, roughly two hours per card, and real expertise. Because Tangem cards cannot be updated, the weakness cannot be patched. Kurennykh’s rebuttal, published on the company’s blog, was that this is precisely the wrong thing to worry about: “LFI attacks are not scalable, and for everyday users, the practical risk is virtually non-existent,” he said, adding that “there have been no known real-world losses from laser fault-injection attacks on any hardware wallet to date.”

Donjon ran a similar attack on the Trezor Safe 7’s TROPIC01 element, disclosed in June, bypassing signature verification on chip samples but defeating only one of several independent security layers, again with no user funds at risk. Žák’s response leaned on the layered design, as crypto.news reported: “Because the Trezor Safe 7 was built with multiple independent security layers, a vulnerability in TROPIC01 does not put user funds at risk.”

Hold the two stories side by side. Breaking the chip took a quarter-million-dollar laboratory, physical possession, hours of work, and produced zero victims. Reading the screen took a cheap implant hidden in a reseller’s box and produced tens of millions in losses. The industry spends its credibility arguing about the attack that does not happen to ordinary people and, until this autumn, under-weighted the one that does.

Security Model, Head to Head

The specification sheets converge on the chip and diverge on everything that surrounds it. The table below lines up the three designs on the attributes that decided who lost money in 2026.

Signing: The Other Thing That Drains Wallets

Chips and seeds aside, the moment a hardware wallet earns its keep is when it shows you what you are about to sign. If the device can only display an unreadable blob of data, which is blind signing, you are approving a transaction on faith, and faith is what drained Bybit and a long list of others. Clear signing, which renders a transaction into human terms (this contract, this amount, this recipient), is the fix, and the standard behind it, ERC-7730, matured enough in 2026 that Ledger handed its stewardship to the Ethereum Foundation in May.

Ledger’s own chief technology officer, Charles Guillemet, has warned for years that a raw transaction is “not intelligible by default. It’s a digital payload,” and that the only safe posture is to “don’t trust, verify.” All three brands can show basic Bitcoin and simple token transfers in clear terms; the gap widens with complex DeFi interactions, where a large screen (Ledger’s Flex and Stax, Trezor’s Safe 5 and 7) can render far more detail than a tap-to-sign card. If most of what you do is approve intricate smart-contract calls, the quality of the signing screen matters more than the brand on it, a cost many traders only see after the fact, as HOGE Wire found when it added up what a perp DEX trade really costs.

The Threats No Chip Stops

A fair comparison has to end outside the device, because that is where most 2026 losses actually happened. Three categories never touch the secure element.

The first is physical coercion, the “$5 wrench attack.” Chainalysis counted about $30 million in funds taken through violent, in-person attacks in the first half of 2026 across 46 known incidents, with kidnappings in a majority of them and France a notable hotspot, in its mid-year report. No chip helps here. Jameson Lopp, who maintains a public database of physical Bitcoin attacks, offers the only durable defense, and it is behavioral: “The most effective thing that a Bitcoiner can do to reduce their wrench attack risk is very difficult: Don’t talk about Bitcoin, at least not while using your real name or face.”

The second is the vendor’s own data. The implant wave rode on top of a year of breaches that handed attackers names, addresses, and phone numbers: Ledger’s 2020 leak (now the subject of that $500 million suit), Trezor’s 81,000-customer ShipMonk breach, and a SafePal breach in August 2026 that exposed order data for close to 40,000 customers. None of that touched a private key. All of it is fuel for the phishing and doorstep attacks that do.

The third is the device’s own randomness. The year’s largest single hardware-wallet loss had nothing to do with a laser or an implant: a Coldcard firmware build flag from 2021 quietly weakened the random-number generator on some devices, cutting effective entropy from 128 bits to as little as 40, and attackers swept roughly 1,816 BTC, about $116 million, from thousands of addresses once they realized the seeds were guessable, TRM Labs found. A later fix cannot heal a seed already generated weakly. The lesson for a buyer is uncomfortable: even the number your wallet picks in secret can be the flaw.

So Which One Should You Actually Buy?

There is no single winner, which is why four of these comparisons can run on this site without repeating an answer. The right device falls out of two questions: what are you actually afraid of, and how much operational discipline will you really practice? Match the threat model to the design.

Note first what self-custody takes off the table and what it puts on. Holding your own keys removes the counterparty risk that an exchange fails or freezes, and it sits outside the perimeter the SEC polices: the regulator oversees exchanges, broker-dealers, and issuers, not the phrase in your pocket, as its own retail-custody guidance makes clear. That independence is the point, and the catch. There is no help desk and no clawback; the chain of custody, the backup, and the signing are all on you. HOGE Wire’s running coverage of where that perimeter is being drawn lives in SEC Crypto Enforcement in 2026.

With that said, here is the matching, by holder.

Two caveats cut across the table. If your portfolio sprawls across many chains and long-tail tokens, including assets whose price keeps moving long after the team behind them has gone quiet, confirm coin support on the exact device before buying, since the thinnest cards and the Bitcoin-only signers will not cover everything a DeFi user touches. HOGE Wire has written about how strange that long tail can get in the case of a dead AI fund whose token still trades.

And whichever logo you choose, the CryptoBilis implants reduced the most important rule to something no brand can sell you: control how the box reaches you, generate your own secret where no screen or camera can see it, and assume the supply chain is hostile until you have checked it yourself. The chip was never the weak link. The path to your door was.

Frequently Asked Questions

Are Ledger and Trezor wallets safe to buy after the CryptoBilis implants?

Yes, if you control how you buy them. The implants were installed by a tampered reseller, not at the factory, and they did not break either company’s secure element. Buy directly from Ledger or Trezor, or from a retailer you can verify; refuse any device that arrives already set up or ships with a recovery phrase; and generate a fresh key yourself on the device. If you bought from CryptoBilis since March 2026, move your funds to a new wallet with a new recovery phrase.

Does Tangem’s seedless design make it the most secure wallet?

It is the most resistant to the specific attack that hit CryptoBilis buyers, because no recovery phrase is displayed for an implant to read unless you turn on BIP39 mode. It is not universally safest: its firmware cannot be updated to patch a future flaw, your key depends entirely on the physical cards you hold, and a lab laser attack has been demonstrated against the chip, though with no real-world losses. Seedless trades the risk of a leaked phrase for the risk of losing every card.

Ledger vs Trezor vs Tangem: which is best for a beginner?

For a first-time holder with a modest balance who worries most about losing a paper backup, Tangem’s tap-to-sign card is the simplest to use safely and removes the seed-phrase step entirely. Readers who want a cheap, auditable, seed-based device often choose a Trezor Safe 3 or a Ledger Nano S Plus. The more important choice for a beginner is buying direct and practicing recovery before funding the wallet.

Is open-source (Trezor) more secure than Ledger’s closed system?

Open-source lets independent researchers find and disclose flaws, and Trezor argues that transparency is itself security. Ledger argues that a certified, closed secure element denies attackers a map. Both positions have merit, and 2026 showed the debate is partly beside the point: the CryptoBilis implant defeated neither chip, it read the screen. Source model matters less than chain of custody and signing quality for most users.

What is the single best way to avoid losing crypto from a hardware wallet?

Control the whole path, not just the chip. Buy direct, verify the tamper-evidence, generate your own seed in private, back it up offline, and use clear signing so you can read every transaction before approving it. Most 2026 losses came from the supply chain, data breaches, weak randomness, phishing, or physical coercion, none of which a secure element can stop on its own.

Yuki Tanaka covers wallets, custody, and self-custody security for HOGE Wire.