Attack ongoing · Updated August 2, 2026

Is My Coldcard Affected?

A flaw in how Coldcard generated seed phrases shipped in firmware from March 2021 to July 2026. Roughly 1,367 BTC has been drained from about 4,585 addresses so far, and the sweeps are still running.

Answer four questions below and find out whether your seed is at risk. The checker never asks for your seed phrase.

Is My Coldcard Affected?

Four questions, none of which are secret. You get a straight answer and the exact steps for your situation.

This tool will never ask for your seed phrase, and nothing that does is legitimate. Fake “checker” sites that ask you to type your twelve or twenty-four words are the standard follow-on scam after an incident like this. Never enter your seed on any website, including this one. Everything below runs in your browser and nothing you select is sent anywhere.

The model name is printed on the back of the device.

Pick your model to start. Nothing you select leaves your browser.

The mistake that will cost people money this week

Plenty of people are going to update their firmware, or import their existing seed words into a brand new hardware wallet, and believe they are done. They are not done, because neither of those actions helps.

The weakness lives in the seed itself, not in the device holding it. Those twelve or twenty-four words were guessable the moment they were created. Typing them into a Jade, a Passport, a Trezor, or anything else carries the same guessable seed to a new box. The only thing that helps is generating a completely new seed and moving your bitcoin to it.

What Happened

A hardware wallet has one job that matters more than every other job combined. It has to generate a seed phrase that nobody can guess. That requires real randomness, and real randomness is why these devices ship with a dedicated hardware random number generator.

Coldcard stopped using its hardware random number generator in March 2021. Nobody noticed for five years and four months.

On July 30, 2026, somebody noticed. An attacker swept thousands of addresses in under an hour. Coinkite published an advisory and shipped emergency firmware the next day.

The Timeline

March 1, 2021
A code change moves seed generation from the call that reaches the hardware random number generator to one that silently resolves to MicroPython’s software fallback. The change ships in firmware 4.0.1.
March 2021 to July 2026
Every seed generated on affected firmware is produced by a predictable software generator seeded from the chip’s serial number and timer registers. The source code is public the entire time. Nobody catches it.
July 30, 2026
An attacker begins draining Coldcard-generated addresses. Reported totals climb from 594 BTC on July 31 to roughly 1,367 BTC across about 4,585 addresses by August 2, near $89 million. The sweeps have not stopped.
July 31, 2026
Coinkite publishes its advisory and ships patched firmware for every affected model. The advisory states the company was unaware of the bug until that day.

What Actually Broke

Coldcard's build defined a configuration flag for the hardware random number generator and set it to zero, because Coinkite supplied its own wrapper instead. A library in the build checked whether that flag existed rather than whether it was turned on. The flag existed, so the check passed, and seed generation was wired to MicroPython's fallback generator.

That fallback was seeded once from the chip's unique ID and its timer registers, and it gathered no fresh entropy afterward. Both of those inputs are guessable. A Bitcoin seed is supposed to carry 128 bits of unguessable randomness. These carried far less.

The obvious question is why the Mk4, Mk5, and Q are affected at all, since those models have secure elements that produce genuine randomness. Block's engineers traced that too. Those devices do mix secure-element entropy back in, but the reseed hashes it and then keeps only four bytes of the result, replacing a single 32-bit word of the broken generator's state. Forty bytes of real randomness go in and 32 bits come out the other side, which caps the whole thing at about 232distinguishable output streams no matter how good the secure element is.

128
bits required
~72
bits on Mk4, Mk5, Q
~40
bits on Mk2, Mk3

Effective search space estimates published by Coinkite. Forty bits is trivially brute-forceable. Seventy-two bits is expensive but reachable, and somebody reached it.

Why I Delisted Coldcard Entirely

Let me be straight with you about one thing first, because I would rather you trust this page than agree with it. Coinkite patched the bug. A Coldcard bought today, running current firmware, generates a proper seed. I am not telling you the device on the shelf is broken right now.

I am telling you not to buy it anyway, and here is why.

Coldcard was sold on the promise of paranoia. Dual secure elements, duress PINs, a self-destruct PIN, air-gapped signing, verifiable source code. That entire feature list is downstream of one assumption: that the seed on the device is unguessable. For five years it was not, and every one of those features protected a key that an attacker could derive without ever touching the hardware.

The part I cannot get past is that the code was public the whole time. Coldcard's pitch was that you did not have to trust Coinkite because you could verify the firmware yourself. Verification is only worth something if somebody actually verifies. Nobody did, for five years, on the most important function in the device.

So my call here rests on the process that shipped the product rather than on the unit in the box. I put a lot of people onto this wallet, and I am not going to keep an affiliate link pointed at it while that call stands.

If You Own a Coldcard, Do This Now

Read this part carefully. Updating your firmware does not fix a seed that was already created on the broken version. The bad randomness happened once, when the seed was born. A patched device cannot go back and un-guess it.

1
Assume you are affected
If you generated your seed on a Coldcard at any point from March 2021 onward, treat it as compromised until you confirm otherwise. Do not wait to be certain.
2
Update the firmware first
Mk2 and Mk3 need 4.2.0 or later. Mk4 and Mk5 need 5.6.0 or later. The Coldcard Q needs 1.5.0Q or later. Edge builds need 6.6.0X, or 6.6.0QX on the Q. You need patched firmware before you can generate a safe replacement seed.
3
Generate a brand new seed on a device you trust
A different hardware wallet is the cleaner choice here, and it is what I would do. If you generate the replacement on the same Coldcard, do it only after the firmware update completes.
4
Understand the front-running risk before you broadcast
The attacker holds the same key you do. The moment your rescue transaction hits the public mempool, anyone watching can see the spend and try to replace it with a higher-fee transaction paying themselves. Several security commentators have raised this; there is no confirmed case of it happening yet, and analysis of the fee-bump activity during the first sweep concluded those were owners raising their own fees rather than attackers racing them. Treat it as a real risk with an unconfirmed body count.
5
Send a small test transaction
Move a small amount to the new wallet, confirm it arrives, and confirm you can spend from it before you move anything else.
6
Move everything else
Sweep the remaining balance to the new wallet, then retire the old seed permanently. Never reuse it for anything.

Two narrow exceptions

Coinkite says a seed is not considered at risk if you added at least 50 fair, independent, private dice rolls when you created it, and those rolls were never recorded or exposed. Your own dice entropy was mixed in, and that part was never broken.

A BIP-39 passphrase is the second exception, and the experts do not fully agree on how much it buys you. Coinkite treats a strong, unique passphrase as protection for the funds behind it, because the passphrase is not derived from the device. Wizardsardine, who published an independent technical analysis, argues you should treat your passphrase as insecure regardless, because typical passphrases carry far less entropy than a seed and BIP-39 applies very little key stretching on top.

I side with the more cautious reading, because a passphrase may buy you time without repairing the seed underneath it. Migrate either way.

Not affected

Coinkite states that TAPSIGNER, OPENDIME, and SATSCARD are not affected by this bug.

The Uncomfortable Lesson

Coinkite's own read on how this got found is worth sitting with. Their position is that somebody likely pointed AI at old firmware releases and let it hunt. Coinkite CEO NVK put it this way: “AI-assisted code review can now find latent bugs at a speed that is outpacing even the industry’s most seasoned experts. If your firmware is open-source or has ever been public, assume it’s already being read by attackers and defenders alike.”

Every hardware wallet with public source code is now being read this way, by people who want to fix what they find and by people who want to spend it. Five years of published firmware is a corpus a machine can chew through in an afternoon.

This does not change my position on self-custody by a single degree. An exchange can lose your bitcoin through incompetence, fraud, or a court order, and you get a claim number instead of your money. A hardware wallet bug is a real risk, and it is still the smaller one. What it changes is how much weight I put on a company's marketing about how paranoid they are.

Sources

Everything on this page comes from Coinkite's own advisory or from reporting that cites it. Read the primary sources yourself.

Common Questions

Is my Coldcard affected?
If the seed was generated on a Mk2 or Mk3 running firmware 4.0.1 through 4.1.9, on a Mk4 or Mk5 running anything before 5.6.0, or on a Coldcard Q running anything before 1.5.0Q, then yes, unless you added at least 50 of your own private dice rolls when you created it. The Mk1 cannot run the affected firmware, so it is out of scope. Use the checker at the top of this page for an answer specific to your setup.
Which Coldcard firmware versions are safe?
Mk2 and Mk3 are fixed in 4.2.0 and later. Mk4 and Mk5 are fixed in 5.6.0 and later. The Coldcard Q is fixed in 1.5.0Q and later. Edge builds are fixed in 6.6.0X, or 6.6.0QX on the Q. All of these shipped on July 31, 2026. What matters is which firmware was running when the seed was first generated, not what the device runs today.
Does updating the firmware fix my seed?
No. This is the most important and most misunderstood point. The firmware update only makes future seeds safe. A seed that was already created by the broken generator stays weak forever. You have to generate a completely new seed and move your bitcoin to it.
Can I just move my seed to a different hardware wallet?
No, and this is the mistake that will cost people money. The weakness is in the seed itself, not in the device holding it. Importing the same twelve or twenty-four words into a Jade, a Passport, a Trezor, or any other wallet carries the weak seed with it. You need a brand new seed generated by something that was never affected, then a transfer of funds to that new wallet.
Does a BIP-39 passphrase protect me?
Partially, and there is genuine disagreement about how much. Coinkite treats a strong, unique passphrase as protection for the funds behind it. Wizardsardine, in an independent analysis, argues you should treat the passphrase as insufficient, because typical passphrases carry far less entropy than a seed and BIP-39 does very little key stretching. I side with the more cautious reading, because a passphrase may buy you time without repairing the seed underneath it. Migrate anyway.
I used dice rolls. Am I safe?
If you added at least 50 fair, independent, private dice rolls during seed creation, Coinkite states the dice input alone contributed at least 128 bits of entropy, which is full strength. The broken generator could not weaken that. This only holds if the rolls were genuinely random and were never recorded or observed. If any of that is uncertain, treat the seed as at risk.
How much bitcoin was stolen?
The figure has climbed as analysts traced more addresses. Early reports on July 31 said 594 BTC. By August 2, reporting put it near 1,367 BTC across roughly 4,585 addresses, worth about $89 million. The attack is still ongoing, so this number is not final.
Are Coldcard checker tools safe to use?
Only if they never ask for your seed phrase. Fake checker sites that ask you to type your twelve or twenty-four words are the standard follow-on scam after a wallet incident, and they exist right now. No legitimate tool needs your seed to answer this question, because the answer depends only on your model, your firmware, and whether you used dice. The checker on this page asks for exactly those three things, runs entirely in your browser, and sends nothing anywhere.
Are TAPSIGNER, OPENDIME, or SATSCARD affected?
No. Coinkite states these products are outside the scope of this bug.
Why not just buy a Coldcard with the patched firmware?
You can. The patched firmware generates proper seeds. I delisted Coldcard because the flaw sat in the single most important function of the device for over five years without anyone catching it, and because the source code was public that entire time. My call rests on the process that produced the device. I make no claim that a new unit is currently broken.
What should I buy instead?
The Blockstream Jade Plus at $149, the Foundation Passport at $199, or the BitBox02 Bitcoin-Only at $150. All three are Bitcoin-only and open-source. Blockstream has published confirmation that Jade is unaffected by this bug.

Buy One of These Instead

All three are Bitcoin-only and open-source, with full reviews on each.

Last updated August 2, 2026. I previously reviewed and recommended the Coldcard Mk4, Mk5, and Q, and I earned affiliate commissions on them. Those reviews are gone and so are the links. If any fact on this page is wrong, message me and I will correct it.