Your seed phrase never left the screen. That is exactly how it was stolen.

—
I have been in this business long enough to have a drawer of hardware wallets I no longer trust, and a smaller drawer of ones I still do. The rule I used for years was simple, and I taught it to people: keep the keys offline, sign on the device, and you have removed the entire class of attacks that eats software wallets. It was a good rule. It was also, it turns out, a rule about cryptography — and cryptography was never the part that broke.
On October 9, 2026, reports started circulating that people were losing funds from Ledger hardware wallets. By October 13, Ledger’s security research team, Donjon, had published its analysis of the attack, and the details are worth reading slowly, because they are not what most people assumed.
Here is what Donjon found, in their own framing: the implant sits inside the enclosure and reads the data shown on the screen. During setup, the signer displays the 24-word recovery phrase. When the implant detects those screens, its microcontroller decodes the words and transmits them over LTE using a provisioned eSIM. And then the sentence that matters most: the implant is a separate component. It does not modify the Secure Element or the Ledger operating system, and it does not alter transactions or signatures.
Read that again. The chip that holds your keys was never touched. The firmware was never touched. Your transactions were never touched. The attacker did not break the cryptographic model of the device at all. They put a second, tiny computer in the box, gave it a cellular radio, and had it watch the one moment in the entire lifecycle of a hardware wallet when the secret is displayed in plaintext on a screen.
Here is the attack as a sequence, stripped of everything that is not load-bearing:
1. device enters an authorized reseller channel intact
2. implant is added inside the enclosure -- a separate component
3. eSIM is provisioned -- the device now has a radio it should not have
4. user unboxes, pairs, updates. every check passes. device signs correctly.
5. setup screen displays the 24-word phrase in plaintext, exactly once
6. implant detects the screen, decodes the words, transmits over LTE
7. attacker now holds a permanent master key. they can wait years.
Note what is absent from that list. There is no step where the Secure Element is defeated, no step where the firmware is modified, and no step where a signature is forged. The device’s security model was not bypassed. It was simply never engaged, because the secret it exists to protect had already left the building by a route the model does not cover.
So this is not a story about broken encryption. It is a story about a broken channel. And the channel is the part nobody audits.
The Failure Chain, Link by Link
When something like this happens, the instinct is to ask “what was the vulnerability?” That is the wrong question. The right question is: how many separate things had to fail, or simply not exist, for a tampered device to end up generating a fresh wallet in someone’s home? Let me walk the chain. There are six links, and only one of them is a technical control.
| # | Link in the chain | Guard that was supposed to hold | What actually held it |
|---|---|---|---|
| 1 | Factory to authorized distributor | Nothing | Nothing |
| 2 | Distributor to reseller | “Authorized reseller” status | A commercial relationship, not a technical control |
| 3 | Reseller shelf to sealed box | Tamper-evident packaging | Re-packaging |
| 4 | Sealed box to user’s desk | User inspection on unboxing | Most users cannot tell an implant from a capacitor |
| 5 | Setup screen | None | The 24 words must be shown in plaintext exactly once |
| 6 | Phrase to attacker | None | The device has no radio. The implant brought its own |
Look at link 5. There is no guard. There cannot be. A hardware wallet has to display the recovery phrase once, in plaintext, so a human can write it down. That is not a bug. That is the product working as designed. Which means the attack does not need to defeat the device’s security — it only needs to be present and watching during a step the device is required to perform.
And link 6 is the part I keep coming back to. The device has no cellular radio. The implant brought its own. An air-gapped signing device, the entire selling point of the category, sitting in a house, quietly holding an LTE session open to a server somewhere. Nothing in the threat model of a hardware wallet includes “the wallet itself makes outbound network connections,” because the wallet is not supposed to have a modem. The attacker did not find a hole in the model. They changed the model.

Why “Authorized Reseller” Is Not a Security Property
CryptoBilis was not a sketchy marketplace listing. It was listed as an authorized reseller in Indonesia, Malaysia, and the Philippines. That word, authorized, is doing a lot of work in this sentence, and almost none of it is technical.
Here is the thing about an authorization list: it is a promise about a business relationship, made at a point in time, enforced by contract rather than by cryptography. It tells you that at some moment a manufacturer believed a distributor would behave. It does not tell you that a device in a box on a shelf in Jakarta came from that distributor. Once a device is legitimately in the channel, the channel itself is the attack surface — and the channel has no signature.
"Authorized reseller" = a commercial assertion about a past state of a business relationship "Verified device" = a cryptographic assertion about a present state of a specific piece of silicon The first is what the industry sells. The second is what users think they are buying.
The economics make this worse rather than better. A reseller with authorization has inventory, shelf space, and customer trust — all of which are worth more than the margin on one tampered unit. The attack is not “sell fake devices.” The attack is “sell real devices, in a real channel, with one extra component.” Nothing about the transaction looks wrong. The receipt is real. The packaging is real. The device boots, pairs, updates, and signs correctly. Every check a user can actually perform passes.
And the checks users are told to perform are the wrong ones. Trezor, in its own public statement, noted that its existing controls can detect counterfeit devices and modified firmware — but that the Ledger implant operated differently, by reading information displayed on the screen. That is the honest version of the problem. Counterfeit detection and firmware verification are checks against a different attack than the one that actually happened.
The Blast Radius Is Bigger Than the Loss Figure
The loss estimates are all over the place, and Ledger has been admirably straight about this: the company says it has not confirmed the total number of affected users or the total value lost. Blockchain investigator Specter put it above $86 million. Bitquery’s figure is above $93 million. Another firm, Yfarmx, went to $103 million. When the spread between estimates is $17 million and the vendor has confirmed neither number, the honest thing to say is that the loss is large and unverified, and to treat any single figure as a headline rather than a measurement.
But the number that actually worries me is a different one: Ledger has sold more than 8 million hardware wallets. Against that base, a loss figure in the tens of millions is a small fraction of users. The point is not the percentage. The point is that the failure mode is not bounded by the users who bought from the bad channel in the known window.
Here is why. The implant lifts the recovery phrase during setup. A recovery phrase is not scoped to one wallet, or one chain, or one moment. It is the master secret for everything that phrase has ever controlled, and everything it will ever control. Someone who set up a tampered device in March, moved their coins to a fresh self-custody wallet in June, and felt safe in October, is not safe. The theft is not of a balance. It is of a key. The attacker does not need to be fast, and they do not need to be clever about timing. They can wait years and sweep at leisure, because the phrase does not expire.
And the channel contamination is spreading beyond the original reseller. MediaMarkt Saturn, a major European electronics retailer, pulled Ledger products from stores and online internationally as of October 13, after a customer in Germany reported an implant — in that case, with no loss of funds. Separately, an X user posted photographs of two Ledger devices bought from two different MediaMarkt stores, reporting unauthorized cables and cellular modules in both. Former Mt. Gox CEO Mark Karpelès said a Ledger Flex purchased at Singapore’s airport contained a spy implant. Ledger has not confirmed whether the European and Singapore reports are linked to the CryptoBilis operation. That is the correct, careful answer — and it is also the answer that tells you the scope is not yet known.

What the Industry Actually Did, and What It Reveals
The responses are instructive, because they are all about the channel and none of them are about the cryptography.
CryptoBilis suspended sales and shipments of all hardware wallet brands across Malaysia, the Philippines, and Indonesia, closed its physical stores, and said it was reviewing its processes with independent specialists. Trezor removed CryptoBilis from its authorized network and, in an open letter on October 13, announced it was revoking authorization for selected third-party resellers effective January 1, 2027, with shipments to those partners stopping immediately — while allowing remaining stock to be sold through Black Friday and the holiday season. Trezor’s chief commercial officer, Danny Sanders, was explicit: there is not a single piece of evidence that Trezor devices have been compromised. OneKey revoked the reseller entirely and offered free replacements for devices bought through CryptoBilis between March 1 and October 12, with applications open through December 31.
Notice the shape of every one of those responses. Shrink the channel. Buy direct. Verify the source. Not one of them is “we improved the device.” Because the device was never the problem. The industry’s answer to a supply-chain attack is a supply-chain remedy, which is honest, and also an admission that the category’s security model has a component that no amount of firmware auditing can reach.
There is a deeper design question here that I do not think the industry has answered. A hardware wallet’s promise is that the keys never touch a networked computer. The implant breaks that promise without touching the wallet. Any defense that lives inside the device enclosure is defending the wrong perimeter. The perimeter that was breached is the perimeter between the factory and your desk — a stretch of road, a warehouse, a shelf, and a courier.
What Actually Works
So what do you do? Not “throw away your wallet.” That is panic, and panic is its own failure mode. But there is a short list of things that are actually load-bearing, and they are all about the channel rather than the crypto.
- Buy from the manufacturer, not from a marketplace, and not from a reseller you cannot audit. This is the single highest-value change. Trezor’s own public guidance is to purchase directly from the manufacturer. The convenience of a local shop or a familiar listing is real, and it is smaller than the risk.
- Treat any third-party device as unverified until you have verified it. If you already own a wallet bought through a reseller, the question is not “is it compromised” but “can I establish that it is not.” Ledger published an inspection guide, with the honest caveat that opening a device can damage it — and the sensible instruction to back up your recovery phrase first, before you open anything.
- Move to a multisig setup with hardware from different manufacturers. This is the one structural fix, and it is the one I would push hardest. Trezor’s guidance explicitly recommends multisignature wallets using hardware from different providers, and the reason is arithmetic: a supply-chain compromise of one vendor’s channel gives an attacker one signature. A 2-of-3 across three vendors from three channels gives them one third of a key. That is not a policy improvement. That is a different security model.
- Understand what your recovery phrase actually is. It is not a password for a wallet. It is a permanent, unscoped master key for every asset it has ever controlled and ever will. If a phrase may have been exposed during setup, the correct response is to generate a new one on a trusted device and move everything — not to move the coins and hope.
- Watch the network behavior of your own hardware. This is the one that sounds paranoid and is not. A hardware wallet has no reason to hold an LTE session. If your device has a cellular module you did not put there, that is not a feature.
The uncomfortable summary is this. For fifteen years, the industry’s security story was “we beat the software wallet attack surface.” That story was true, and it was incomplete. It described the cryptography and said nothing about the logistics. The Secure Element did its job perfectly. It protected a key that had already been photographed on its way out of the screen.
The next generation of hardware wallet security is not going to be a better chip. It is going to be a signed chain of custody from the factory to your hands, and a threat model that finally admits the device is not the only computer in the box.
(The End)


