Coldcard, made by the Canadian firm Coinkite, is one of the most established names in bitcoin-only hardware wallets. Its entire design philosophy rests on one idea: private keys should be generated and kept air-gapped, offline, and beyond the reach of an internet-connected computer. That discipline is the reason the device earned trust in the self-custody community in the first place.
Late in July 2026, that trust was tested. Coinkite disclosed a firmware defect, present since March 2021, that in some cases replaced true hardware-generated randomness with a predictable substitute. Instead of drawing on the device's dedicated random number generator, affected firmware fell back to a software routine seeded with non-secret data - the chip's serial number and its internal clock. On certain devices, this reduced the effective entropy of a wallet seed from an expected 128 bits down toward levels a determined attacker can search computationally, rather than one that requires breaking real cryptography.
Coinkite has said an internal review did not catch the issue before it became public. It surfaced only after an attacker found and used it.
Coldcard Exploit: What Happened, in Sequence
The exploitation played out in at least three distinct waves. An initial sweep on July 30 emptied hundreds of single-signature wallets in a window of roughly 25 to 40 minutes, moving several hundred BTC. Follow-up waves through July 31 and August 1 extended the pattern to thousands more addresses, and a fourth wave the following week pulled several hundred more BTC from wallets that hadn't yet migrated. Independent trackers, including Galaxy Research, have tied the activity to thousands of affected addresses and losses that have been revised upward more than once since the first disclosure - most tallies now sit in the range of roughly $90–100 million.
The affected coins span the full life of the flaw. Seeds created any time from 2021 through mid-2026 on vulnerable firmware are exposed, meaning both freshly funded wallets and long-dormant ones have been at risk simultaneously.
Why a Firmware Update Alone Doesn't Fix It
This is the detail that matters most for anyone who owns an affected device. The vulnerability lived in *how a seed was originally generated*, not in the firmware currently running. Installing a patch stops the device from generating new weak seeds going forward - it does nothing to repair a seed that already exists. Coinkite's own guidance is blunt on this point: *"An update is not a seed migration."*
Anyone who generated a seed on an exposed configuration needs to create an entirely new seed on patched firmware and move funds to it, rather than assume an update alone resolves the exposure. Coinkite has named the affected configurations directly: Mk2 and Mk3 devices on firmware 4.0.1 through 4.1.9, and Mk4, Mk5, and Q devices on any version prior to their respective fixed releases. Migration carries its own risk, since the transaction that moves funds off a compromised wallet still has to be broadcast from that same compromised key - a detail that has pushed some researchers to recommend private or out-of-band broadcasting services during the transition. TAPSIGNER, OPENDIME, and SATSCARD run on separate codebases and are not affected.
What This Isn't
Every other major hardware wallet manufacturer - Ledger, Trezor, BitBox among them - has stated its products are not affected. This is not an industry-wide flaw in how hardware wallets work; it is specific to a build error in one company's firmware, where a check for whether the device's true hardware random number generator was enabled failed silently and fell through to a much weaker software fallback.
That distinction matters. Self-custody as a principle is not what failed here. A single implementation did.
The Pattern Behind the Failure
Coldcard's firmware is open source, and has been the entire time this flaw existed. That fact deserves sitting with. Open availability of code is necessary for verifiability, but it is not sufficient for security - a defect can sit in plain sight for years without anyone auditing it closely enough to notice. Several researchers investigating the incident have suggested the attacker may have used automated tooling, including AI-assisted code review, to find the weakness faster than any human audit had. Coinkite has said its own AI-assisted review of the same codebase, conducted shortly before the exploit, missed it too.
This is not the first time a security failure has reshaped how the bitcoin industry operates, and it won't be the last. Exchange collapses forced a rethink of custodial cold storage practice. Earlier key-management bugs forced better standards around seed generation and hardware entropy sourcing. The lesson each time has been the same: the companies that treat independent, adversarial auditing as a continuous baseline - not a one-time certification or a line in a pitch deck - are the ones that hold up. The ones that treat it as optional pay for that choice publicly, and their users pay first.
Update, August 20: Firmware 5.6.1 Rebuilds the Entropy Model
Three weeks after the initial disclosure, Coinkite shipped its most substantial response yet: firmware 5.6.1 for the Mk4 and Mk5, and 1.5.1Q for the Q, alongside the earlier 4.2.0 release for Mk2/Mk3 and 6.6.0X/6.6.0QX for Edge users. This is not a cosmetic patch. It rebuilds the seed-generation pipeline from the ground up, and it changes the practical advice every owner should be following.
A hardened entropy source. Master seed generation now draws fresh entropy from three independent sources at once - the device's STM32 hardware RNG and both of its Secure Elements - rather than relying on a single path that a build error could quietly bypass. The random-number engine itself has moved to a NIST-standard Hash-DRBG construction, replacing the older PRNG implicated in the original flaw, and the device now runs a boot-time self-test that halts startup entirely if the hardware entropy path can't be verified. In plain terms: the class of failure that caused the July breach - a silent fallback to predictable data - is now checked for and blocked at the hardware level, not just patched around.
User entropy is no longer optional. This is the change most owners will notice directly. Every newly generated master seed, temporary seed, and multisig co-signing key now *requires* the user to contribute physical entropy - dice rolls, coin flips, or timed key-press mashing - on top of the device's own randomness. The device checks that input for obviously non-random patterns before accepting it. It's a meaningful philosophical shift: rather than asking users to trust the hardware's randomness alone, the new standard mixes verifiable human-supplied entropy in as a mandatory ingredient, not an advanced, opt-in workflow reserved for the paranoid.
The passphrase, formalized as a security layer. Coinkite's own advisory carries a notable detail that has been true throughout this incident: seeds protected by a strong, unique BIP-39 passphrase were never exposed by the underlying RNG bug, regardless of firmware version, because a passphrase-protected wallet derives from more than device-generated entropy alone. That fact has effectively promoted the passphrase from a power-user feature to a recommended baseline. The 1.5.1Q release goes a step further on the interface side, now requiring users to scroll through a locally entered BIP-39 passphrase before it's revealed on screen - a deliberate friction point designed to prevent a passphrase from being glimpsed by anyone standing nearby during entry.
What this means in practice. If you're setting up a new Coldcard - or migrating an old seed onto patched firmware, as every affected owner needs to do - the current guidance is to pair the new seed with a strong, unique passphrase from the outset, rather than treating it as an optional extra layered on later. Coinkite's documentation is explicit that a lost or mistyped passphrase cannot be recovered, so it needs to be backed up exactly and separately from the seed itself, with its wallet fingerprint recorded and the full recovery path tested before any real funds are deposited. Used this way, the passphrase does double duty: it hardens a wallet against exactly the kind of device-level entropy failure this incident exposed, and it adds a second, independent factor that a thief who somehow obtains the physical seed backup still can't spend without.
What Coldcard Owners Should Do
- Do not assume a firmware update resolves your exposure. It only prevents new insecure seeds from being generated.
- If your seed was generated on an affected model and firmware version, generate a brand-new seed on patched firmware and move your funds to it.
- Set a strong, unique BIP-39 passphrase on the new seed. Back it up exactly and separately from the seed words, record its wallet fingerprint, and test the full recovery path before depositing real funds. Treat this as standard practice going forward, not an advanced option.
- Consider a multisignature setup. A flaw in a single device's entropy source is far less catastrophic when no single device holds a complete, spendable key.
- Verify Coinkite's official advisory directly before acting, since exposure depends on your specific device model and the firmware version active when your seed was first created.
Final Note
Hardware wallets exist to remove trust in any single third party. This incident is a reminder that removing trust in an exchange or a custodian doesn't remove the need for trust somewhere - it moves that trust into the device, the firmware, and the audit process behind it. The August update is the industry's answer to that problem in miniature: more entropy sources instead of one, mandatory human-supplied randomness instead of optional, and a passphrase treated as a default line of defense rather than an afterthought. Evaluating a hardware wallet on those terms - layered, verifiable, and assuming failure rather than promising invulnerability - is the discipline this event should reinforce across the industry.
House of Bitcoin covers bitcoin infrastructure and security through evidence-based research. This piece reflects Coinkite's public disclosures through August 23, 2026, and will be updated if the scope of the incident or the firmware guidance changes.
