Coldcard Hack: Is Your Bitcoin Safe? The $116M Flaw That Drained 5,200 Wallets — 7 Checks Before You Trust Your Hardware Wallet

On July 30, 2026, attackers started draining Bitcoin from wallets that were supposed to be unhackable — and by August 5 they had taken roughly 1,816 BTC, worth about $116 million, from more than 5,200 addresses. The wallets were Coldcards, the hardware devices Bitcoin's most security-conscious holders trust above all others. The flaw had sat in the firmware since March 2021, passed more than 15 security reviews, and was only found by a new AI model. If you own a Coldcard — or any hardware wallet — the seven checks in this article decide whether your coins are safe right now.

Coldcard is made by the Canadian company Coinkite, and for years it has been the wallet Bitcoin purists recommend for the safest possible self-custody. It is air-gapped, open-source, and built around a dedicated chip whose only job is producing unpredictable randomness for seed phrases. That reputation is exactly why this story matters: nobody broke into a house, phished a password, or compromised a device. Attackers reconstructed private keys with maths alone, offline, because the seeds themselves were weaker than anyone knew. Updating your firmware does not fix it — a weak seed stays weak forever.

What Happened: The $116 Million Drain in Four Waves

The thefts began on July 30, 2026. According to analysis by TRM Labs, citing Galaxy Research's running tally, the first wave alone drained 1,196 addresses in 41 minutes — roughly $70 million at the time. Three more waves followed, and by August 5 the preliminary total stood at approximately 1,816 BTC, then worth nearly $116 million, taken from more than 5,200 addresses. About 88% of the compromised Bitcoin had sat untouched for at least a year — the victims were mostly long-term holders.

The attackers did not stop there. Public research has since identified at least 15 separate attacker groups targeting the same flaw, and Coinkite has described the threat as ongoing. Galaxy Research tracked one attacker swapping roughly 10% of its haul through THORChain into Ether. This was not one lucky hacker. It was a structural weakness, and once the method was understood, every weak seed became a target.

Date	What happened	Scale
March 2021	Firmware change routes seed generation through a weak software random number generator	Flaw introduced, undetected
July 30, 2026	First drain wave begins	1,196 addresses in 41 minutes, ~$70M
August 1, 2026	Coinkite widens its advisory beyond the Mk3 line	All models confirmed affected
August 5, 2026	TRM Labs / Galaxy tally published	~1,816 BTC (~$116M), 5,200+ addresses
November 9, 2026	Coinkite CEO defends the security reviews; developers dispute him	Debate over audits reignites

The timeline matters because of what it says about detection. The bug lived in publicly available, open-source firmware for more than five years. It survived two paid third-party product reviews, a review using OpenAI's Cybersecurity Harness that produced 85 candidate findings, and at least 15 documented external reviews touching Coldcard hardware, firmware, random-number-generation tools, or release controls. None of them found it. That is the detail that should change how every crypto holder thinks about the word "audited."

The Flaw: A 2021 Firmware Change That Bypassed the Security Chip

A Bitcoin seed phrase — the 12 or 24 words that control every coin in a wallet — is only as safe as the randomness used to create it. A proper seed draws on 128 bits of entropy, meaning an attacker would need to try roughly 340 undecillion combinations to guess it. Coldcard was designed to generate that randomness in a dedicated hardware chip, the STM32 true random number generator, precisely so that no software bug could weaken it.

In March 2021, a single firmware change broke that design. The seed generation call was switched from the hardware random number generator to a software library called libngu, which routed through MicroPython's Yasmarang pseudo-random number generator — a non-cryptographic generator never meant for security-critical work. The device kept working normally — the seeds still looked random, but they were not random enough.

Coinkite's own investigation estimated that affected Mk2 and Mk3 wallets produced seeds with only about 40 bits of effective security, while Mk4, Mk5, and Q wallets produced roughly 72 bits, against the intended 128. Dropping from 128 bits to 40 bits shrinks the search space from an effectively impossible task to roughly one trillion possibilities — a number a well-resourced attacker can chew through with computation alone. That is how the coins were stolen: the private keys were regenerated offline, without ever touching a device, a PIN, or a password.

The distinction between the models matters, but not in the way early reports suggested. Coinkite's initial advisory called the newer models unaffected, but its own follow-up and an independent analysis by Block showed that Mk4, Mk5, and Q devices share the underlying fallback and reach only about 72 bits of effective entropy. Seventy-two bits is far better than 40, but it is still far below the 128-bit standard. One researcher's public reconstruction of the first wave recovered 328 wallet seeds behind 1,042 of the 1,195 source addresses.

The Numbers That Decide If Your Coins Are at Risk

Number	What it means	Why it matters to you
$116 million	Estimated value stolen by August 5, 2026	The largest hardware-wallet seed theft ever recorded
1,816 BTC	Coins drained from 5,200+ addresses in four waves	The scale of the exposure, not a single incident
40 bits	Effective seed strength on Mk2/Mk3 (vs 128 intended)	Weak enough to brute-force with computation alone
72 bits	Effective seed strength on Mk4/Mk5/Q	Better, but still far below the security standard
5 years	How long the flaw sat in public firmware	Any seed made between March 2021 and July 2026 may be affected

The window is the critical fact. Any seed generated on Coldcard firmware released between March 2021 (version 4.0.0) and the July 31, 2026 fix is potentially affected — Mk3 devices running firmware 4.0.1 through 4.1.9, and Mk4, Mk5, or Q devices running firmware before the fixed releases. Fixed firmware is available for every supported model: Mk2/Mk3 version 4.2.0, Mk4/Mk5 5.6.3, Q 1.5.3Q, plus Edge releases 6.6.0X and 6.6.0QX. Installing it only protects seeds created after the update — it does nothing for a seed that already exists.

The November 9 Fight: "It Took an AI Model to Find What Humans Couldn't"

The story came roaring back on November 9, 2026, when Coinkite CEO Rodolfo Novak — known as NVK — publicly defended the security reviews Coldcard had undergone. At least 15 external reviews or research efforts had touched Coldcard before the defect emerged, including two paid third-party product reviews, and a review by Rob Ham using OpenAI's Cybersecurity Harness that produced 85 candidate findings and prompted several fixes. The seed-generation defect was absent from all of them. "The uncomfortable truth is that it took a new AI model to find what humans couldn't," NVK wrote.

Bitcoin developers pushed back immediately. Luke Dashjr disputed the claim that human reviewers could not have identified the problem, arguing that competent review of the random-number-generator code could have caught it. James O'Beirne said he had raised concerns about Coldcard's randomness implementation as early as May 2025. Both sides can be partly right: an audit is a snapshot of specific code, at a specific time, against a specific scope — Coinkite itself now distinguishes targeted independent checks from a complete audit of every corrected firmware binary. A device can be open-source, audited, and air-gapped — and still carry a five-year-old flaw.

What Most People Get Wrong About Hardware Wallet Safety

Myth one — "Audited means safe." Coldcard underwent at least 15 reviews and two paid third-party audits before the flaw was found. Audits reduce risk; they do not eliminate it. The BitGo analysis of the incident makes the deeper point: the real failure was a single point of failure in how seeds were generated, and the fix is redundancy — multi-signature setups and independent key generation — not a better audit badge.

Myth three — "Updating the firmware fixes it." This is the most dangerous misunderstanding. Coinkite's advisory states it plainly: updating corrects future seed generation but does not repair an existing affected seed. The weakness lives in the number the machine already picked.

Myth five — "Only Coldcard users are affected." The flaw is Coldcard-specific, but the lesson is not. Every hardware wallet depends on a random number generator, and every one of them could have a similar failure mode. That is why the industry response has centred on multi-signature wallets and independent entropy. See our guide to how to store crypto safely in 2026 for the full background.

The 7 Checks: Is Your Coldcard Safe?

Work through these in order. If any check fails, treat the wallet as exposed and move to the migration steps in the next section.

• Check your seed date first. If your seed was generated on Coldcard firmware released between March 2021 and July 31, 2026, assume it is affected. Seeds made before March 2021 or after the fixed firmware releases are outside the window. If you cannot remember when you created the wallet, assume the worst — that is the safe default, and it costs nothing.
• Check your model and firmware version. Open Settings on the device and compare the release against Coinkite's advisory. Mk3 devices on firmware 4.0.1 through 4.1.9 are in the most severely affected range, at roughly 40 bits. Mk4, Mk5, and Q devices on firmware before 5.6.0 and 1.5.0Q respectively are affected at roughly 72 bits. Install the current recommended release for your model — but remember that this step protects future seeds only.
• Check whether you added dice rolls when you created the seed. Coinkite's advisory gives a specific exception: seeds created with at least 50 fair, independent, private dice rolls, with a sequence that was never recorded or exposed, are not considered vulnerable to this flaw alone. If you rolled dice during setup, you are probably safe. If you let the device generate the seed on its own, you are not.
• Check whether you use a strong, unique BIP-39 passphrase. A passphrase of 20 or more characters, never typed into a computer, adds a separate barrier that the flaw alone does not defeat. It does not repair an affected seed, but an attacker who reconstructs the seed still cannot spend the coins without the passphrase. If you never set one, this protection does not exist for your wallet.
• Check whether your wallet is part of a multi-signature setup. If your quorum includes keys generated on non-Coldcard devices, the attacker would need to compromise those keys too, and the Coldcard flaw alone is not enough. This is the single strongest structural defence, and it is the reason BitGo's analysis recommends multi-signature custody as the lesson of the incident.
• Check your balances and recent activity now. Look at every address the wallet controls, including old and dormant ones. The attackers targeted precisely the wallets nobody was watching, and the thefts happened in waves over days. If funds are gone, report it — recovery depends on law enforcement and exchange freezes, not a refund.
• Check whether anyone else holds a copy of your seed. The flaw is about randomness, not theft of the paper backup, but the incident is a reminder that a seed's security is only as strong as its weakest copy. If your seed words were ever photographed, typed into a computer, or stored in a cloud note, the firmware question is no longer your biggest problem. Our comparison of the best crypto hardware wallets in 2026 covers the storage practices that matter.

What To Do Right Now If You Own a Coldcard

If any check above failed, do not panic — but do not wait. The exploit is still active, with at least 15 attacker groups reported to be targeting the same flaw. Coinkite's open letter put it bluntly: if you generated a seed using a Coldcard wallet, move your funds now. Here is the migration sequence from the official guidance:

1. Verify your existing backup first. Confirm you can still access your current seed words and passphrase first.
2. Install the fixed firmware for your model. Download it only from Coinkite's official channels. Mk4/Mk5 should run 5.6.3, Q should run 1.5.3Q, and Mk2/Mk3 should run 4.2.0, the final legacy release for that hardware line.
3. Generate a completely new seed on the patched firmware. Use at least 50 physical dice rolls for extra entropy, set a strong, unique BIP-39 passphrase, and record the new wallet fingerprint.
4. Verify a receiving address on the device screen. Never trust an address shown only on a computer screen. Confirm it on the Coldcard's own display before sending anything.
5. Send a small test transaction first. Move a tiny amount, wait for confirmation, and confirm it arrived in the new wallet before moving the rest.
6. Move all remaining funds. Every funded balance, including dust and old addresses.
7. Never restore the old seed anywhere. Importing it into a new device, a software wallet, or an updated Coldcard carries the weakness forward. The old seed is retired, permanently.

One more warning: expect phishing. After every major hack, scammers impersonate the affected company and send fake "urgent migration" links. Coinkite will never ask for your seed words, and no legitimate migration requires typing them into a website. The same pattern played out after the NGRAVE phishing attack earlier this year — our breakdown of the NGRAVE hack and the $2.56 million phishing theft shows exactly how these follow-up scams work, and the hardware wallet safety guide covers the physical and social defences that matter too.

Coldcard vs the Alternatives After the Hack

Does this incident mean Coldcard is finished, or that another brand is automatically safer? Neither. Every hardware wallet is a bet on one vendor's random number generator, firmware process, and disclosure culture. The honest comparison is about structure, not brand loyalty.

Setup	Single point of failure	What the Coldcard flaw would have done	Verdict after the hack
Single Coldcard, device-generated seed	The seed itself	Full loss — keys reconstructable offline	Migrate now if your seed is from the affected window
Single Coldcard, 50+ dice rolls	None from this flaw	Nothing — independent entropy defeats the bug	Safe from this flaw; still a single point of failure generally
Coldcard + BIP-39 passphrase	The passphrase	Nothing without the passphrase	Strong mitigation, but the seed is still weak underneath
Multi-signature (Coldcard + other vendors)	No single device	Nothing — quorum needs the other keys too	The strongest structure; the incident's main lesson
Ledger or Trezor, single signature	That vendor's RNG	Not affected by this flaw — but exposed to their own equivalent	Different risk, same shape; add dice rolls and a passphrase
Exchange custody	The exchange	Not affected — but see the $6.34B Binance heist	You trade device risk for counterparty risk

Frequently Asked Questions

Is my Coldcard still safe to use?

The device itself was never hacked or remotely accessed — the flaw was in how some seeds were generated, not in the hardware. Fixed firmware generates proper seeds going forward. But if your existing seed was created on affected firmware between March 2021 and July 2026 without dice rolls, the device is safe and the seed is not — move your funds to a brand-new seed.

How do I know if my seed was affected?

Seeds generated on firmware from March 2021 to July 31, 2026 are in the affected window: Mk3 on firmware 4.0.1–4.1.9, and Mk4, Mk5, or Q on firmware before the fixed releases. You are likely safe if you used at least 50 dice rolls during creation or a strong unique BIP-39 passphrase. If you are unsure, treat the seed as affected and migrate — the cost of migrating is far lower than the cost of being wrong.

Does updating the firmware fix my wallet?

No. Updating the firmware corrects seed generation for new wallets, but it does not change or repair an existing seed. The weakness is in the number already chosen. The only fix is to generate a completely new seed on patched firmware, with at least 50 dice rolls, and move all funds to it. Restoring the old seed on any device carries the weakness forward.

What is the $116 million figure based on?

TRM Labs' analysis of August 5, 2026, citing Galaxy Research's running tally, put the thefts at approximately 1,816 BTC from more than 5,200 addresses across four waves beginning July 30, worth nearly $116 million at the time. Later research identified at least 15 attacker groups targeting the same flaw, so the total exposure is likely larger than the initial tally.

Are Ledger and Trezor affected by the same flaw?

No — the flaw is specific to Coldcard firmware from Coinkite. But the structural lesson applies to every hardware wallet: each one depends on its vendor's randomness implementation. Dice rolls, a strong passphrase, and multi-signature setups protect you regardless of vendor.

What should I do if my Bitcoin was already stolen?

Stolen Bitcoin is traceable on the public blockchain, and firms like TRM Labs and Galaxy Research have tracked the Coldcard thefts. Report the theft to law enforcement with your transaction records — but beware of recovery scams, which surge after every major hack and target victims a second time. No one can reverse a confirmed Bitcoin transaction; recovery depends on exchanges freezing the funds if the attackers try to cash out.

Conclusion

The Coldcard hack is the most important hardware wallet story of 2026 because it broke the assumption the entire self-custody movement rests on: that the device you trust to generate your keys does so correctly. A five-year-old firmware bug quietly cut seed security from 128 bits to as little as 40 and cost holders roughly $116 million before anyone noticed.

But the story is not a reason to abandon self-custody. It is a reason to finish it properly. The holders who lost nothing were the ones who added their own randomness with dice rolls, set a strong passphrase, or spread their keys across a multi-signature setup — three defences that each cost almost nothing and each, on its own, would have stopped this attack. If you own a Coldcard, run the seven checks today, and if any of them fail, migrate to a fresh seed on fixed firmware. Bitcoin at roughly $75,000 is volatile enough without holding it in a wallet whose keys can be guessed. The device did not fail you — the assumption did.