Atlas

Ledger SAS · Nano S+, Nano X, Flex, Stax

Ledger

An ST33 with a vendor OS (BOLOS): keys are born and sign inside the chip. This is not an authenticator, and Recover is possible only because Ledger can program the silicon.

Vendor-OS SEMulti-coinApproachableRecover

The ST33 is not an ATECC608. It is a certified small computer, with an OS (BOLOS), applications signed by Ledger, field updates. Keys are born and sign inside the chip; the interface MCU displays, talks USB or Bluetooth, and never sees the seed. Attestation says “this is a genuine Ledger”. It does not give you a BOLOS hash you compiled.

The Nano S+ and X remain small-screen dongles. Flex and Stax widen the e-ink, a necessary — not sufficient — condition for clear signing: read the recipient on the device, not on the PC.

Recover, a paid 2023 option, exfiltrates nothing until you enable it. It showed something else: a vendor-OS SE can learn to shard entropy. A fixed-instruction authenticator would not have learned it. The debate is not “closed source equals backdoor”. It is “who is allowed to load code into the chip that holds the seed”.

Fits if

Multi-coin beginners, daily use, resistance to physical theft.

Keep in mind

Vendor trust model + closed SE OS. Bluetooth and USB widen the surface. Recover is optional; programmability of the ST33 is not.

Related incidents

What Recover does, and what it broke in the story

Recover is not a seed theft. It is not proof that “closed source equals backdoor” either. It is an optional, paid, identified service that splits entropy 2-of-3 across three custodians. It exists because a vendor-OS SE can learn new tricks via update — an ATECC608 cannot. The anger of 2023 came from a slogan: a capsule whose key never left. What showed up was a programmable computer.

The protocol, step by step

  1. 01

    You subscribe — nothing is automatic

    Ledger Recover is a paid, identified (KYC) subscription. Until you activate it, the flow described here does not start. The device works as before: seed on the SE, paper or steel backup on you. The 2023 debate was not “Ledger activated Recover on every Nano”, it was “Ledger showed that the SE firmware could learn to export fragments”.

  2. 02

    Entropy is encrypted inside the Secure Element

    It is not your 24-word phrase that travels in the clear. The SE takes the entropy (the binary matter of which the words are only a transcription), encrypts it, then splits it. The custodians do not receive “word1 word2 word3…”. Each receives an encrypted fragment, useless in isolation, and which is not a copy of the seed.

  3. 03

    Pedersen Verifiable Secret Sharing, 2-of-3

    Ledger uses a Shamir variant called Pedersen Verifiable Secret Sharing (PVSS). Like Shamir, it lets you reconstruct a secret once you hold a threshold of shares — here two of three — and a single share reveals nothing about the secret. “Verifiable” adds that you can check the consistency of the shares without reassembling them. This is honest cryptography, not a marketing sleight of hand.

  4. 04

    Three custodians, two are enough

    The three fragments go to three distinct custodians: Ledger, Coincover, and EscrowTech (the trio announced at launch; commercial names may have changed, the 2-of-3 scheme has not). Each custodian stores its share in an HSM. A corrupted or seized custodian has nothing spendable. Two custodians who collude, or who are compelled together, can reconstruct. That is the real trust threshold: not one, two.

  5. 05

    Reconstruction goes to a device, after identity

    To recover, you prove your identity (the original KYC), you get two fragments sent back to a Ledger, and the SE reassembles. The stated goal is to save people who lost their paper — a real problem, statistically more common than glitching. The cost is introducing companies, an identity process, and an export path that did not exist in the original Nano story.

Myths worth killing

“Ledger can read my seed whenever it wants.”

No, not with the announced scheme. One share is not enough. Ledger alone, without Coincover or EscrowTech, does not reconstruct. On the other hand, Ledger wrote the firmware that knows how to fragment and export, and Ledger is one of the three custodians. This is not “they have the seed”. It is “they designed a path, and they hold one key out of three”.

“If I don’t activate Recover, I am out of reach.”

For Recover as sold, yes: no subscription, no fragments at the custodians. The friction point is elsewhere. Closed silicon, updated by the manufacturer, could in theory receive other instructions. Ledger states that the export exists only in the Recover flow, optional. You cannot verify that yourself. That is exactly the “company + certification” trust model, not the “I read the C” model.

“Recover sends my seed to Ledger.”

The protocol sends encrypted entropy fragments, not the word list. Linguistically, “sharing the seed” is therefore false. Conceptually, sharing what it takes to reconstruct it at 2-of-3 is indeed sharing the power to recreate the vault. Both sentences can be said in bad faith. Here we say the third: we share a reconstruction power, under a threshold, with KYC.

“It’s the same thing as SLIP39 on Trezor.”

No. With SLIP39, you choose the custodians (you, your sister, a safe). No KYC, no subscription, no company holding a share by default. With Recover, the custodians are companies, identity is bound, the service can stop, be seized, or refuse a country. The threshold cryptography looks similar. The politics do not.

“Closed source = less safe.”

Not automatically. A well-certified closed SE has held physical attacks that open MCUs have not. Recover did not “break” the ST33. It changed what people thought Ledger allowed itself to make the chip do. Practical safety and trust model are two axes. We keep them apart on purpose.

Questions still asked

You want Shamir without a company

SLIP39 on Trezor, SeedXOR on Coldcard, or a multi-vendor 2-of-3. You pick the custodians — not Ledger, Coincover or EscrowTech.

Seed, Shamir, steel

Other vaults

Trezor Safe 3 / 5

Open MCU, opaque silicon

Trezor Safe 7

Open MCU, opaque silicon

Coldcard

Open MCU, opaque silicon

BitBox

Open MCU, opaque silicon