Why Certified Secure Element Chips Couldn't Stop the Hardware Wallet Supply-Chain Attack Used Against Ledger Hardware Wallets

In October 2026, attackers stole an estimated $86–93 million from more than 300 crypto wallets without ever breaking a secure element. They opened Ledger devices sold through a single reseller, added a cellular implant beside the certified chip, and read each owner's 24-word recovery phrase off the screen during setup. The incident shows that a certified chip does not make a certified device. Ledger and Trezor both use high-grade, Common Criteria–certified secure elements, but the device around the chip (screen, battery, microcontroller, radio, housing) sits outside any certification boundary. And the most valuable secret, the recovery phrase, is designed to leave the chip and be shown to a human.

By Sue Pontius, Chief Executive Officer, Lokblok · Published 19 October 2026

The Lokblok Toughkey Phantom Secrets Answer: No Standing Secrets

The Toughkey Phantom Secrets answer removes the target rather than guarding it harder:

  • No recovery phrase exists. There is nothing to display, write down, photograph or transmit.

  • The Protected Key never persists. It is reconstructed only momentarily inside certified hardware, used, and then destroyed.

  • The secure element is the security boundary. Toughkey runs on NXP JCOP 4.5 (Java Card 3.0.5) on the P71D600, a platform validated to FIPS 140-3 Level 3 with Level 4 physical security, and certified to Common Criteria EAL6+.

  • No single device is enough. Recovery requires a quorum of independent Recovery Agents, so one tampered unit does not yield the key. Further, members of the quorum will not know each other so they cannot conspire with each other.

What happened

All confirmed losses trace to Ledger devices sold by CryptoBilis, a Malaysian reseller; Ledger has suspended its sales. Thefts began around 9 October 2026.

How the implant worked. Researchers found the original battery swapped for a smaller one to make room for extra circuitry: an STM32 microcontroller, a 2×2 mm eSIM and a shielded LTE module. Ledger said the implant read the 24-word recovery phrase from the screen and sent it over a mobile network. It did not modify the secure element or Ledger OS.

Why the checks missed it. Ledger's Genuine Check verifies the secure element. Ledger acknowledges it cannot detect added hardware while the original secure element is intact. CEO Pascal Gauthier called physical supply-chain tampering a risk for the whole industry.

Trezor's response. On 13 October, Trezor announced it will cut its reseller network, with affected authorizations ending 1 January 2027. Trezor states there is no evidence that any Trezor device was compromised and describes the move as preventive. CryptoBilis also sold Trezor devices and was removed from Trezor's list on 11 October.

The significance is that both market leaders are now treating the distribution channel as part of the attack surface. Trezor's own letter to resellers makes the point: a hardware wallet is only as safe as the path it travels to reach its owner.

The real lesson: the certified boundary ends at the chip

The attack succeeded because the secret that matters most was designed to leave the certified chip. A chip certification covers the chip. It says nothing about the screen the phrase is drawn on, the microcontroller that drives it, the battery bay, or the housing a reseller can open.

A hardware wallet built this way has two security boundaries:

  1. The secure element — certified, tamper-responsive, holds the keys.

  2. The device — uncertified. It includes the display, general-purpose MCU, radios, power and enclosure, and it sees the recovery phrase in clear text at setup.

An attacker does not need to break boundary 1 if boundary 2 shows them the answer. That is exactly what happened: the secure element was untouched, the Genuine Check passed, and the funds were gone.

This is why the question is not only which chip a device uses but where the secret is allowed to exist. Any design that turns the master secret into 24 human-readable words creates a moment when it sits outside every certified boundary. A certified chip, FIPS or Common Criteria, cannot protect a secret it has already handed to the screen.

How Ledger and Trezor are built

Both vendors use strong secure elements, certified under Common Criteria rather than validated under FIPS 140-3. Neither markets a FIPS 140-3 validation for the chip or for the finished device.

Ledger. The Nano X uses a STMicroelectronics ST33J2M0 secure element certified to CC EAL5+, running Ledger's own BOLOS operating system. The device also contains a general-purpose MCU, Bluetooth radio, battery and display outside the secure element.

Trezor. The Safe 3 and Safe 5 use the Infineon OPTIGA Trust M (V3), certified to CC EAL6+; the Safe 7 adds Tropic Square's TROPIC01. In the Safe 3 and Safe 5, the secure element protects the PIN and releases a secret that decrypts the wallet backup, which is stored on the general-purpose chip. The open-source firmware runs on that separate MCU.

What both share. Each generates a BIP-39 recovery phrase, shows it on a screen driven outside the certified chip, and asks the owner to keep it on paper or metal. Each device is assembled from certified and uncertified parts inside a housing that is not itself a certified boundary. That is a reasonable design for consumer self-custody; the CryptoBilis case shows its limit when the supply chain is compromised.

Note: Trezor reports no tampered Trezor devices. The point here is architectural, not an allegation against Trezor products.

The Lokblok Toughkey Phantom Secrets answer

Toughkey™ answers the CryptoBilis attack by never creating the secret the implant was built to steal. Phantom Secrets™ keeps every sensitive operation inside the certified chip and leaves nothing of value outside it.

1. The certified chip is the security boundary

Toughkey runs Lokblok's own firmware on NXP JCOP 4.5, a Java Card 3.0.5 and GlobalPlatform operating system, on the P71D600 secure element. That platform carries:

  • FIPS 140-3, overall Level 3, physical security Level 4 — CMVP certificate #4679, active to March 2029.

  • Common Criteria EAL6+ — NSCIB certification of JCOP 4.5 P71.

  • Active tamper response — per NXP's security policy, a metal shield mesh over the die, voltage, temperature, light and glitch sensors, and zeroization on detected intrusion. Secrets in non-volatile memory are encrypted under an on-chip master key.

Phantom Secrets uses the chip's validated primitives directly: ECDH key agreement (KAS-ECC SP 800-56A), HKDF key derivation (SP 800-56C), the approved DRBG, AES and ECDSA. These are the same building blocks the Phantom Secrets Security Model rests on.

2. There is NO recovery phrase

Phantom Secrets replaces the seed phrase with a threshold construction in which no secret persists between operations.

  • Protected Key — the key being protected. It is never stored at rest. It is reconstructed momentarily inside the secure element, used, and then immediately destroyed.

  • Recovery Agent Credential — held privately by each Recovery Agent, sealed inside that agent's Toughkey. It never leaves the chip. It does not contain or store any part of the Protected Key; it only lets that agent participate in reconstruction.

  • Regen Token — the only persistent artifact. It is public, can be stored anywhere, and on its own reveals nothing about the Protected Key.

At recovery, each of a quorum of Recovery Agents re-derives a private value from its own credential, pairs it with its public Regen Token, and the Protected Key is momentarily reconstructed inside certified hardware, then again destroyed. The key owner authenticates with ToughID™ biometrics.

3. What a CryptoBilis-style implant would find

Apply the same attack to a Toughkey: open the housing, add an MCU, an eSIM and an LTE module, and resell it.

  • No phrase on any screen. There are no 24 words to capture at setup, ever.

  • No exportable secret on the bus. The Recovery Agent Credential and the reconstructed Protected Key exist only inside the P71D600. The implant can watch traffic to the chip but not the secrets processed within it.

  • Regen Tokens are public by design. Capturing them gives the attacker nothing they could not already be handed.

  • One device is not enough. Even full control of one Toughkey does not satisfy the quorum; recovery needs independent Recovery Agents on separate devices.

4. Form factors, Secure Terminal and attestation

Toughkey ships as a USB key, a smart card, or a cloud HSM. The USB key and smart card have no screen and no battery. Besides the secure element, the USB key carries only a USB/NFC interface chip that relays commands and responses; it never handles a secret. Every secret lives inside the secure element's own package, validated to FIPS 140-3 Level 3 with Level 4 physical security: it detects tampering and zeroizes its secrets in response. The cloud HSMs Toughkey uses are also validated to FIPS 140-3 Level 3.

Every form factor works with Lokblok Secure Terminal. Secure Terminal pairs a web page, a WebAssembly component and a policy suite. A local sidecar service enforces that policy, with default-deny access, and brokers every call to the Toughkey. The page never gets direct access to the device. Secure Terminal runs in the operating system's native webview, not a consumer browser.

This is where Phantom Secrets changes the outcome. The Protected Key is reconstructed inside the secure element, used, and burned there. It never crosses the USB or NFC interface, so an implant on the interface chip would see only commands, results and public Regen Tokens. A Ledger implant had a 24-word phrase to steal; a Toughkey implant has nothing.

Toughkey also uses hardware attestation at enrollment and on every key function. Attestation validates chain of custody: before any operation, the platform confirms it is talking to a genuine Toughkey secure element. Ledger's Genuine Check could make the same confirmation and still lose funds, because its secret left the chip. Toughkey's never does.

The attacker's prize, the thing that made $86–93 million portable over a mobile network, does not exist. As Lokblok puts it: you cannot steal what is not there.

Backup without a recovery phrase

The CryptoBilis implant targeted one moment: when a BIP-39 wallet shows its owner the words they must write down. That moment exists only because BIP-39 is the backup. Phantom Secrets replaces BIP-39 as the backup mechanism, so the moment never comes.

How a BIP-39 backup works

BIP-39 encodes the master secret as 12 or 24 words. The words are the wallet: anyone who holds them can rebuild it on any device, with no PIN and no second party.

  • It must be shown to be useful. The device has to display the words so a person can copy them. That display is the point the implant read.

  • Every copy is a complete copy. Paper, steel plate, photo, a note in a safe or with an heir — each one is the whole key.

  • No quorum, no revocation. One stolen copy is enough, and a copy cannot be cancelled once made.

How a Phantom Secrets quorum backup works

A Phantom Secrets backup has two parts, and neither one is the key:

  • Regen Tokens — public, storable anywhere, useless alone.

  • Recovery Agents — independent parties, each holding a Recovery Agent Credential sealed in their own Toughkey.

Recovery requires a quorum of Recovery Agents working with their Regen Tokens. The Protected Key is reconstructed inside the secure element, used, and burned. Nothing is written down, nothing is displayed, and no single item, person or device can recover the key alone.

Two ways a key enters Toughkey

Generated on board. The key is created inside the secure element and backed up to Phantom Secrets at once. It is never displayed and never exported, because the quorum backup means there is never a reason to export it. This path has no exposure point from creation onward.

Imported. An existing key or BIP-39 phrase, such as a treasury wallet being migrated, has to be entered once. If it is typed on a keyboard or shown on any display during import, that single moment is exposed, as is every copy that existed before. After import, the key is backed up in Phantom Secrets and never needs to be displayed or exported again.

For the highest assurance, generate the key on board and move funds to it. When importing is necessary, treat the old phrase as retired: destroy remaining copies once the Phantom Secrets backup is confirmed.

BIP-39 recovery phrase Phantom Secrets quorum backup
What is stored The secret itself, as words Public Regen Tokens
Shown to a person Yes, at creation Never (generated on board)
Enough to steal one item? Yes No — quorum required
Where the secret is rebuilt Any device, anywhere Inside a certified secure element only

Side by side

Ledger Nano X Trezor Safe 3 / 5 Lokblok Toughkey Phantom Secrets
Secure element ST33J2M0 Infineon OPTIGA Trust M (V3) NXP P71D600
Chip certification CC EAL5+ CC EAL6+ CC EAL6+; FIPS 140-3 L3 platform (cert #4679), physical L4
Operating environment Ledger BOLOS Open-source firmware on separate MCU JCOP 4.5, Java Card 3.0.5, GlobalPlatform
Master secret format 24-word recovery phrase Recovery phrase (BIP-39 / SLIP-39) None — Protected Key reconstructed transiently
Secret shown on a screen Yes, at setup Yes, at setup Never
Persistent backup Phrase on paper/metal Phrase on paper/metal Public Regen Tokens only
One device compromised = funds lost? Yes, if phrase captured Yes, if phrase captured No — quorum of Recovery Agents required
Known supply-chain incident CryptoBilis, Oct 2026 None reported None reported

What Toughkey does not claim

A security paper earns trust by stating its limits. These are Toughkey's.

  • The FIPS certificate is NXP's. Certificate #4679 validates JCOP 4.5 on P71D600 with NXP's own applets, as configured in its security policy. It does not by itself validate Lokblok's applet or the finished Toughkey. Toughkey inherits the platform's tamper protections and approved algorithms; a module-level validation of Toughkey is a separate step.

  • Supply-chain risk is industry-wide. Any device can be opened and modified. Phantom Secrets limits what an implant can gain; it does not make tampering impossible. Buying through controlled channels still matters.

  • The host still matters. Phantom Secrets protects keys, not intent. Users should still confirm what they authorize on a trusted display or policy.

  • No chip is unbreakable. Well-resourced labs have attacked secure elements with fault injection and side channels. Level 4 physical security and EAL6+ raise the cost sharply but are not a guarantee.

  • Security rests on stated assumptions. Per the Phantom Secrets Security Model, secrecy depends on independently random polynomial coefficients and on the hardness of ECDH or the chosen post-quantum KEM and the KDF — not on any single certification.

Conclusion

The CryptoBilis thefts were not a failure of secure-element engineering. They were the predictable result of an architecture that hands its master secret to an uncertified screen inside an openable box. Tightening reseller lists, as Ledger and Trezor are now doing, narrows the channel but leaves the target in place.

Lokblok’s Toughkey Phantom Secrets removes the target. The Protected Key lives only for the instant it is needed, inside a FIPS 140-3 Level 3, CC EAL6+ secure element. What persists is public. What is private never leaves the chip. And no single device, tampered or not, can recover the key alone.

About the author

Sue Pontius

Chief Executive Officer, Lokblok

Sue Pontius is CEO of Lokblok, where she leads the company's work on zero-persistence cryptography for digital assets, identity, and high-assurance custody.

View LinkedIn profile →