Trezor Confirms Cellular Implants in Three Safe 5 Wallets Sold by One Reseller, but Two Were Identified From Owners’ Photos, the 2.13.1 Changelog Never Mentions the Fix, and Recovery Words Captured Earlier Stay Exposed
Trezor’s November 12 update says three Safe 5 wallets bought from CryptoBilis between August 20 and September 23 carried a board that copies the screen and sends it over the mobile network; only one...
Trezor says it has now confirmed hardware implants in three Trezor Safe 5 wallets that were bought from the Southeast Asian reseller CryptoBilis. Its November 12 update describes “an extra circuit board fitted between the Trezor Safe 5’s main board and its screen” that copies what the screen shows, recovery words included, and sends it over the mobile network to a server controlled by the attackers. Trezor released firmware 2.13.1 on November 11 with a protection that it says stops the implants it has found from capturing anything, and it adds that the update “can’t protect a wallet backup that was captured before it.” BleepingComputer reported the update on November 13.
Table Of Content
- What Trezor says it found, and what each claim rests on
- The firmware fix: what Trezor claims and what the public record shows
- Ledger first, Trezor second: similar implants by both companies’ accounts, different loss records
- Timeline
- Where the accounts differ
- What we checked ourselves
- What to do now
- What is still unknown
- Sources and method
The update is a confirmation, not a closed case. Only one of the three implants was examined in Trezor’s lab; the other two were identified from photos that their owners sent. The public changelog for 2.13.1 does not mention the protection, Trezor says it is withholding technical details, and its replacement offer reaches back to March 2026 while the three confirmed purchases fall between August 20 and September 23. Below we set out what the update supports, how it compares with Ledger’s earlier account of the same reseller, where the coverage and the sources differ, what we could check ourselves, and what is still unknown.
What Trezor says it found, and what each claim rests on
The table follows the November 12 update. The middle column quotes it or reports it closely. The right column says what the claim rests on, which is our reading of the post.
| Claim | What Trezor says | What it rests on |
|---|---|---|
| The hardware | An extra board between the main board and the screen that “only listens.” | Lab work on the first device: “Our embedded team has extracted and analyzed the firmware from the implant in the reported Trezor Safe 5.” |
| What it captures | It “copies everything shown on the screen, including what you type on the device.” That covers recovery words shown when a backup is created and typed when a wallet is recovered, and a passphrase if one appears on screen. | The same lab analysis. |
| How it sends | It encrypts what it captures and “sends it over the mobile network to a server controlled by the attackers.” If it cannot connect, it “keeps the data and tries again until it succeeds.” | The same analysis. The post names no server and lists no indicator. |
| Which devices | Three Safe 5 units, all “bought between 20 August and 23 September 2026,” through CryptoBilis’s Tokopedia and Shopee stores and directly from CryptoBilis. | One device was examined in the lab. “The other two were identified from photos sent by their owners.” |
| How far it spreads | None found in devices from any other reseller or in any other Trezor model. 48 customers have contacted Trezor, and their devices are being checked “in our lab or from photos of the internals.” | A search that is still running. The post gives no count of Safe 5 units that CryptoBilis sold. |
| Losses | “No customer has reported losing funds.” | Customer reports to Trezor. Trezor says it will update the post if that changes. |
| Link to Ledger | “It works the same way as the implants found in Ledger devices, adapted to the Trezor Safe 5’s display.” | Trezor’s comparison. The post shows no side-by-side analysis. |
Two details limit what “confirmed” covers. First, a photo of a board’s internals can show that an extra board is present, but it cannot show what the board does, so the description of its behavior comes from the one device whose firmware Trezor extracted. That is our reading. Second, on October 30 Trezor said it had “found no signs of tampering on any of the four devices we’ve received so far,” so not every device it examined from the same reseller carried an implant.
The firmware fix: what Trezor claims and what the public record shows
Trezor’s wording is that firmware 2.13.1, “released on November 11th, includes a protection that stops the implants we’ve found from capturing anything on devices that run it.” Two limits come from the same post. The protection applies only to “devices that run it,” and the update cannot protect a backup captured earlier, so Trezor still tells CryptoBilis buyers who set up a device to move their funds to a new wallet with a new seed.
We checked the record around the release.
- The changelog says nothing about it. Trezor’s release feed for the Safe 5 lists 10 changes for 2.13.1: Solana, Ethereum and Cardano items, the removal of Omni support, a fix for characters above U+FFFF being displayed as unrelated characters, and a warning on xpub export. None mentions an implant, tampering or the recovery-word screens. The same items appear in the project’s CHANGELOG.md, with four of them under Security.
- The dates are only partly consistent. The six feed files and firmware binaries we requested, including the older 2.13.0 binary, were all last modified between 12:18:03 and 12:18:08 GMT on Wednesday, November 11. That shows the feed was rewritten that day, which fits Trezor’s date but does not date 2.13.1 on its own. The CHANGELOG.md on the main branch still dates 2.13.1 as “18th November 2026,” a week later, and we do not know why the two differ.
- The source revision is public. The feed’s
firmware_revisionfor the Safe 5 build is4289e23b317e98ab6a5e8237723e0cd57f44934c, the same commit that the core/v2.13.1 tag points to, committed on November 5. We did not rebuild the firmware or compare it with the binary. Trezor says “We won’t publish technical details that could help attackers or put affected customers at risk,” and we are not describing which change carries the protection.
Two questions remain open, and both are our reading. First, “making the current batch of implants useless” assumes that implants already in the field cannot change. Trezor describes a board that reaches the mobile network, and the post does not say whether it tested if the implant’s own firmware can be updated remotely. Second, the post says the implant “never changes anything on the device itself, including the firmware and Secure Element.” By that description nothing the device reports about its own firmware or Secure Element would differ, so a check that only asks the device about itself would not obviously see the board.
Ledger first, Trezor second: similar implants by both companies’ accounts, different loss records
Trezor’s October 30 warning followed Ledger’s CryptoBilis disclosures by three weeks. The table sets Ledger Donjon’s October 13 update beside Trezor’s November 12 account.
| Point | Ledger (October 13) | Trezor (November 12) |
|---|---|---|
| Where the implant sits | “The implant sits inside the enclosure.” | “An extra circuit board fitted between the Trezor Safe 5’s main board and its screen.” |
| What it reads | It “reads the data shown on the screen” and decodes the 24-word recovery phrase. | It “copies everything shown on the screen,” and stores recovery words and passphrases. |
| How it sends | “Transmits them over LTE using a provisioned eSIM.” | “Sends it over the mobile network.” |
| Does it touch the device? | “It does not modify the Secure Element or the Ledger operating system, and it does not alter transactions or signatures.” | It “never changes anything on the device itself, including the firmware and Secure Element.” |
| Losses | “All confirmed fund losses tied to the incident disclosed on October 9th involve signers sold through CryptoBilis,” but “We have not confirmed the total number of affected users or the total value lost.” | “No customer has reported losing funds.” |
Loss figures for the Ledger side come from third parties. Decrypt reported that an on-chain investigator named Specter “traced more than $86 million in suspected thefts across Ethereum, Tron and Bitcoin,” and Ledger says figures circulating online come from third parties and have not been fully verified. The Singapore Police Force’s October 28 advisory counts at least two cases since September 10 with losses of “about $3.5 million,” involving Ledger wallets “purchased from some retail stores in Singapore.” It does not name CryptoBilis, although BleepingComputer’s October 30 report says the police announcement “added another $3,500,000 to the tally.”
The absence of reported Trezor losses is not evidence that nothing was captured. By Trezor’s own description the implant stores the words and sends them to a server, and what the attackers do with them afterwards is a separate step, so a wallet that has not been drained says nothing about whether its phrase was seen. That is our reading. Trezor’s update gives no reason why the two companies’ loss records differ, and we have none. Our August report on a firmware flaw in Coldcard wallets covered a different failure, a defect in the wallet’s own code. Here, by Trezor’s description, the firmware and Secure Element are untouched and the attacker sits outside them.
Timeline
| Date (2026) | Event | Source |
|---|---|---|
| August 20 to September 23 | The three confirmed Safe 5 units are bought, 35 days counting both ends. | Trezor, Nov 12 |
| September 10 | Start of the Singapore police’s count of at least two tampered-Ledger cases. | Singapore Police Force |
| September 16 | BleepingComputer says researchers have shown the implant in a Ledger device “since September 16.” | BleepingComputer, Oct 12 |
| October 9 | Ledger begins investigating CryptoBilis-linked losses. Trezor removes CryptoBilis as a reseller. | Ledger, Trezor |
| October 10 | Ledger confirms an implant in one user’s device. | Ledger, Oct 13 |
| October 28 | Trezor receives a responsible disclosure of one implant in a Safe 5 bought in August. | Trezor, Oct 30 |
| October 30 | Trezor warns CryptoBilis buyers. The device has not arrived, and four devices received so far show no tampering. | Trezor, Oct 30 |
| November 5 | Commit that the 2.13.1 tag points to. | GitHub |
| November 11 | Firmware 2.13.1 is published. | Trezor, release feed |
| November 12 | Trezor reports three confirmed implants, 34 days after it removed CryptoBilis and 13 days after its first warning. | Trezor, Nov 12 |
| November 13 | BleepingComputer reports the update. | BleepingComputer |
Where the accounts differ
- The exposure window. Trezor’s advice and its replacement offer cover anyone who bought from CryptoBilis “from March 2026 onwards,” about seven months before it removed the reseller on October 9. The three confirmed purchases span 35 days in August and September. Trezor’s summary says the implants it has found are limited to devices bought “between August and September 2026.”
- How many devices. Three are confirmed among the devices Trezor has checked. Its pool is the 48 customers who have contacted it, and the post does not say how many devices that covers, whether the original reporter is among the 48, or how many Safe 5 units CryptoBilis sold, so the numbers give no prevalence.
- Where CryptoBilis is. Trezor says only “Southeast Asia.” BleepingComputer’s October 12 report calls it “a major Malaysia-based hardware wallet retailer,” and its November 13 report calls it an “Indonesian reseller.” The October 30 report says the first implant was found by an Indonesian user who bought through Tokopedia. The retailer’s own site prices in ringgit and says orders are shipped “from KL,” which we read as Kuala Lumpur, so the Indonesian label seems to belong to the buyer.
- The cause. Neither Trezor’s two posts nor Ledger’s October 13 update says how or when the boards were fitted. BleepingComputer’s October 12 report said it was “still unclear exactly how or when the devices were tampered with,” and the November 12 update adds nothing on that point. None of the sources we read names an actor.
What we checked ourselves
- Firmware. We read the Safe 5 and Safe 3 release feeds, requested the binaries’ headers, compared the feed’s
firmware_revisionwith the GitHub tag, and read the project changelog. The results are in the section above. - The reseller’s storefront. At about 10:15 GMT on November 15 the store’s Trezor category listed 8 products and its Ledger category 10, and the page markup flags every one of them as out of stock. The Safe 5 is still listed at RM 620 to RM 700 on its product page, and the home page still has Ledger and Trezor menu entries beside the line “All orders are shipped out on the next business day from KL.” We did not place an order or contact the store.
- The 80 percent reseller cut. We could not check it. Trezor gives no before-and-after count, its public reseller page shows 12 entries and a “More resellers” button, and our requests for earlier Internet Archive captures were rate-limited.
- The implant itself. We did not see, test or buy any hardware. Everything about what the board does rests on Trezor and Ledger.
What to do now
This is Trezor’s published guidance, with one addition of ours marked as such.
- Bought from CryptoBilis since March 2026 and already set up. Move funds to a newly created wallet with a new seed. Trezor adds: “Do not reuse the old seed in any wallet, hardware or software.”
- Not set up yet. “Don’t set it up, and never enter your existing wallet backup on it.”
- Replacement. Keep the device and contact Trezor Support. Trezor offers a new device, a different model with a refund of the price difference, delivery to another country, or a refund. For future purchases it recommends “buying directly from trezor.io.”
- Scams. Trezor warns that scammers pose as Trezor and offer replacements or refunds, and says it will never ask for a recovery phrase.
- Our addition. By Trezor’s own limits, the 2.13.1 update helps only on a device that runs it before the recovery words appear. It is not a substitute for replacing a seed that a device has already displayed.
What is still unknown
- How and when the boards were fitted, and by whom.
- How many Safe 5 units CryptoBilis sold, and how many have been checked in a lab rather than from photos.
- Whether the implants can be updated over the mobile network, which decides whether “useless” holds.
- Whether the same board turns up in other resellers’ devices or other Trezor models. Trezor says it will update its post if so.
- Whether any Trezor user has lost funds, and why Ledger’s CryptoBilis devices produced confirmed losses while Trezor reports none.
Sources and method
We read Trezor’s November 12 update and its October 30 post, Trezor’s October 30 thread on X through a public mirror, Ledger’s October 13 post, the Singapore Police Force advisory, Decrypt’s report on Ledger, and BleepingComputer’s stories of November 13, October 30 and October 12. First-hand, we fetched Trezor’s firmware release feeds and the response headers of the firmware files, the GitHub tag and commit records and the project changelog, the reseller page, and the category pages of the reseller’s storefront. The day counts, the 35-day span and the seven-month span are our arithmetic. We did not open or test any hardware wallet, did not buy anything, did not contact the reseller, and did not try to identify or reproduce the implant, so every statement about what it does rests on Trezor’s and Ledger’s accounts.







No Comment! Be the first one.