At 2:14 a.m. UTC on July 31, 2026, a single Bitcoin address began moving. By 2:39 a.m., the sweep was finished. Roughly 594 BTC, worth about $38 million, had left approximately 500 separate wallets and consolidated into one. The victims were not day traders clicking fake links. They were not DeFi degens approving malicious contracts. They were Bitcoin maximalists, security researchers, and long-term holders who did everything right. They bought hardware wallets, kept them air-gapped, never typed their seed phrases into a computer, and stored their devices in safes. They were supposed to be the most secure people in crypto. They were wrong.

The 25-Minute Heist Nobody Saw Coming

The attack did not look like a hack. There were no phishing emails. No compromised exchanges. No social engineering phone calls. The attacker simply knew the private keys. Not because they stole them, but because they could compute them.

Between 01:31 and 01:56 UTC, more than 1,300 individual UTXOs moved across 500 transactions inside a three-block window.

Every drained wallet shared the same profile. Each was single-signature. Each held more than 0.15 BTC. Most had been dormant for years. And critically, every wallet creation date fell between 2021 and 2026. That date range was not random. It matched the exact window during which a firmware bug was live.

The attacker was not watching the blockchain for targets. The targets were determined by when each wallet was created. That is the signature of a seed-generation flaw, not a network breach.

How a Five-Year-Old Code Break Killed the "Unhackable" Myth

Coldcard is manufactured by Coinkite, a Canadian company that has built its reputation on being the hardware wallet for people who do not trust anyone. The devices are bitcoin-only, open-source, and air-gapped. They are the wallet of choice for people who think Ledger's multi-coin approach is too complex and exchange custody is too risky.

The vulnerability began with a single commit on March 1, 2021.

A firmware build error caused Coldcard's seed generation to skip its dedicated hardware random number generator and fall back to a software substitute. The problem sat in a preprocessor guard in the libngu cryptographic library. The guard checked whether a configuration macro existed. It did not check whether the macro was actually enabled.

Because the macro was defined (even though its value was zero), the build system silently linked the software fallback instead of the hardware chip. The build completed without warnings. The seeds looked normal. The addresses accepted deposits. Everything appeared to work. For five years and four months, nobody noticed.

The Entropy Problem: Why 128 Bits Became 40

Here is what that build error actually means. When you create a Bitcoin wallet, your device is supposed to generate a seed phrase from a pool of randomness so vast that guessing it is physically impossible. That is the 128-bit security standard behind BIP-39, the protocol that governs most seed phrases. A 128-bit seed has more possible combinations than there are atoms in the observable universe.

On affected Coldcard Mk3 devices, the effective entropy collapsed to approximately 40 bits.

Think of it this way. A 128-bit seed is a lock with so many possible combinations that every computer on Earth running for a billion years could not try them all. A 40-bit seed is a lock with about one trillion combinations. That sounds like a lot until you realize a single modern graphics card can test billions of combinations per second. A moderately resourced attacker with a few servers can brute force one trillion combinations in a reasonable timeframe.

The Mk4, Q, and Mk5 models included additional secure elements that mixed extra entropy into the process. Coinkite estimates those models produced seeds with roughly 72 bits of effective entropy under the bug.

That is better than 40 bits. It is still roughly 56 bits below the standard. It is the difference between a lock that cannot be picked and a lock that a well-funded adversary can eventually kick open.

The most critical detail in Coinkite's advisory is this single sentence: "Updating the firmware does not change or repair an existing seed."

If your seed was generated on vulnerable firmware, the private key itself is weak. Patching the device does not strengthen a key that already exists. You cannot fix a bad seed with a software update. You must replace the key entirely.

Who Got Hit and Why Dormant Wallets Were the Perfect Targets

The victim profile tells its own story. Every drained wallet was single-signature. Each held more than 0.15 BTC. Most had been inactive for years.

Dormant wallets are ideal targets because their owners are not watching. A day trader checks balances daily. A long-term holder who bought in 2021 and stored the device in a safe might not look at the address for months. By the time they check, the funds are gone and the consolidation address has already moved them through mixers or exchanges.

Galaxy Research estimates total losses may now exceed $70 million across approximately 1,200 addresses.

The initial 594 BTC sweep may be only the most visible portion of a much larger precomputed attack. The attacker spent weeks or months generating candidate keys and matching them to blockchain addresses. The 25-minute on-chain execution was simply the withdrawal phase of a heist that had already been solved offline.

The AI Twist: How the Bug Was Found and Missed

Here is where the story becomes even more unsettling. Coinkite says it ran an AI-assisted audit of its own firmware weeks before the attack. The AI found nothing.

Then the attacker likely used AI to analyze the same open-source codebase and identify the build error.

Several developers were able to reproduce the vulnerability using frontier AI models within hours of disclosure.

This creates a paradox that every hardware wallet manufacturer must now confront. AI can find bugs that human audits miss. But AI can also miss bugs that other AI models catch. The attack surface is no longer just human versus human. It is human plus AI versus human plus different AI, and the defender's AI does not always win.

Are Other Hardware Wallets Safe? The Honest Answer

If you own a Ledger, Trezor, or Bitkey, you are not affected by this specific bug. Block, Trezor, and Ledger have all confirmed their products use different entropy generation approaches.

That does not mean they are invincible. It means they are not vulnerable to this exact build error. The Coldcard incident is not a reason to panic about every hardware wallet. It is a reason to understand that hardware isolation (keeping keys offline) and entropy quality (how random the keys are) are two separate problems. Coldcard failed on entropy quality while maintaining perfect hardware isolation. A device can be air-gapped and still generate predictable keys.

The honest answer is this. No single hardware wallet should be treated as a complete security system. It is one component. For meaningful balances, it should be part of a larger architecture, not the entire architecture.

What You Must Do Tonight If You Own a Coldcard

If you generated a seed on a Coldcard Mk3 running firmware 4.0.1 or later, and you did not add at least 50 independent dice rolls or a strong BIP-39 passphrase during setup, treat your seed as compromised. Not at risk. Compromised. Here is what you do.

First, identify which device model and firmware version were running when you created your seed. Not what is running now. The version at creation. Check Coldcard's historical firmware documentation at blog.coinkite.com if you are unsure.

Second, if you are in the vulnerable group, do not wait for confirmation that you were specifically targeted. The attack cost the attacker nothing per additional target. If your seed was weak, it is only a matter of time before it is tested.

Third, update your device to the latest firmware. Mk3 users should update to 4.2.0 or later. Mk4 and Mk5 users need 5.6.0 or later. Q users need 1.5.0Q or later.

Fourth, generate a completely new seed on the patched device. Verify the backup. Send a small test transaction. Then move your remaining funds. Do not reuse the old seed anywhere.

Fifth, if you exported a seed from a vulnerable Coldcard to another wallet, that seed is still compromised. Moving a weak seed to a secure device does not strengthen it. The weakness is in the seed itself, not the device holding it.

One common mistake is rushing the migration and fat-fingering a transaction. Another is assuming that because your wallet is currently full, you were not targeted. The attacker may simply not have reached your address in their enumeration yet. Speed matters, but accuracy matters more. Move deliberately.

The Bigger Picture: Is Self-Custody Still Worth It?

This incident has reignited a debate that never really went away. For years, the Bitcoin community's answer to exchange risk was simple. Buy a hardware wallet. Write down 24 words. Sleep soundly. FTX, Celsius, and BlockFi made that advice look brilliant. Coldcard just made it look incomplete.

Bitcoin commentator Guy Swann called this "the worst hit in bitcoin history to the most knowledgeable and 'properly secured' bitcoiners."

It was not an exchange getting hacked because of hot keys. It was thousands of individuals having their personal private keys recreated from underneath them.

Some industry observers believe this will accelerate adoption of regulated custodians and spot Bitcoin ETFs.

That is a logical reaction, but it trades one risk for another. ETFs eliminate firmware risk but introduce counterparty risk, regulatory freeze risk, and the inability to move your Bitcoin on a Sunday afternoon without permission.

The better answer is probably the dual strategy that Nick Szabo and others have advocated. Use institutional rails for convenience and scale. But protect the right to self-custody, and practice it with a security model that assumes any single device can fail. Self-custody is not dead. Single-device custody is dead.

The Multi-Sig Solution: Don't Put All Your Trust in One Device

Every wallet drained in this incident was single-signature. That is not a coincidence. A multi-signature setup requiring two of three keys to spend, with each key generated on a different device from a different manufacturer, would have survived this flaw entirely.

If an attacker compromised your Coldcard key, they would still need your Ledger key and your Trezor key to move funds. The computational cost explodes from cracking one weak seed to cracking three separate seeds simultaneously, two of which were generated on unaffected devices.

Kevin Loaec of Wizardsardine, which builds multi-sig custody tools, was among the first to raise the alarm before official advisories circulated.

His message was not to abandon hardware wallets. It was to stop trusting any single one of them as a complete solution.

For balances that would hurt to lose, the new standard should be multi-sig across manufacturers. Not because any particular manufacturer is bad, but because the entire industry just learned that even the most paranoid, open-source, bitcoin-only company can ship a five-year-old entropy bug without detection.

FAQ’s

Q: What is the Coldcard hack?
A: The Coldcard hack was a July 31, 2026 theft of roughly 594 BTC ($38 million) from approximately 500 Bitcoin wallets. Attackers exploited a five-year-old firmware build error that caused certain Coldcard devices to generate weak seed phrases using predictable software randomness instead of hardware randomness.

Q: Does updating Coldcard firmware fix a compromised seed?
A: No. Updating firmware only corrects future seed generation. If your seed was created on vulnerable firmware, the private key itself was generated with weak randomness. The only fix is to generate a completely new seed on patched firmware and migrate your funds.

Q: Which Coldcard models are affected?
A: Mk3 devices running firmware 4.0.1 or later are worst affected with approximately 40 bits of entropy. Mk4, Mk5, and Q devices generated seeds with roughly 72 bits of entropy, still below the 128-bit standard. Tapsigner, Opendime, and Satscard products are unaffected.

Q: Is my Ledger or Trezor safe?
A: Block, Trezor, and Ledger have confirmed their products are unaffected by this specific bug. The vulnerability is specific to Coldcard's firmware build process.

Q: How much Bitcoin was actually stolen?
A: Initial on-chain data shows 594 BTC, worth roughly $38 million. Galaxy Research estimates total losses may exceed $70 million across approximately 1,200 addresses as the situation develops.

Q: What should I do if I own a Coldcard?
A: Identify the firmware version running when you created your seed. If it was an Mk3 on 4.0.1 or later without dice-roll entropy or a passphrase, generate a new seed on patched firmware and migrate funds immediately. Do not reuse the old seed.

Q: Were multi-signature wallets affected?
A: Multi-sig arrangements using only vulnerable Coldcard devices inherit the full vulnerability. However, a multi-sig setup across different manufacturers would have survived, because the attacker would need to compromise multiple independent keys.

Q: How did the attacker find this bug?
A: Coinkite believes the attacker used AI to analyze Coldcard's open-source firmware. Coinkite had run its own AI audit weeks earlier and found nothing.

KEY TAKEAWAYS

• A firmware build error in Coldcard wallets caused seed generation to use weak software randomness instead of hardware randomness for five years.
• Approximately 594 BTC ($38 million) was stolen from 500 wallets in 25 minutes on July 31, 2026. Total losses may exceed $70 million.
• Updating firmware does not repair an existing compromised seed. You must create a new wallet and migrate funds.
• Mk3 devices are worst affected with roughly 40 bits of entropy. Mk4, Mk5, and Q devices are also below standard at approximately 72 bits.
• Multi-signature setups across different manufacturers would have survived this attack. Single-device custody is no longer sufficient for large balances.
• Self-custody remains sound, but the "not your keys, not your coins" mantra now requires a second clause: not your entropy, not your security.

DISCLAIMER

This article is for informational and educational purposes only. It does not constitute financial, investment, or security advice. Cryptocurrency markets and self-custody tools carry significant risks, including total loss of capital. The technical details described reflect publicly available information as of August 2, 2026. Always verify security advisories directly from manufacturers, conduct your own research, and consult qualified professionals before making custody decisions. Past security incidents do not guarantee future vulnerabilities, and no storage method is entirely risk-free.