Somewhere in the early hours of July 30, an attacker drained 1,196 Bitcoin wallets in 41 minutes. No malware. No phishing. No physical contact with a single device. The victims were Coldcard users โ the segment of the Bitcoin community that prides itself on maximum paranoia. The haul: approximately $70 million.
Here is the part that should unsettle every self-custody advocate: this wasn't a zero-day exploit. It was a coding error from March 2021 that silently redirected random number generation to a weak fallback scheme โ one dependent entirely on device serial numbers and a clock. Four years of silent operation. Four years of unknowing exposure. The wallets being swept today were created during that window.
Note: The hardware wallet's security guarantee was never about the chip. It was about entropy.
Coldcard occupies a unique position in Bitcoin's security stack. Manufactured by Canadian firm Coinkite, it is the wallet of choice for bitcoin maximalists who reject the closed-source architectures of Ledger and Trezor. Full offline operation. Open-source firmware. A brand built on the promise that private keys never leave the device. Its users skew sophisticated โ long-term holders with meaningful balances, the kind of people who run their own nodes and dismiss exchange custody as amateur hour.
That sophistication didn't matter. The attack required no interaction with the victim. No malicious link. No fake firmware update. No social engineering. It was a bulk statistical sweep against a compromised key space โ and it worked because the public blockchain makes balance-checking trivially accessible to anyone with a script and a search index. This is also the year crypto theft hit record highs; the market is increasingly numb to headlines about drained wallets. That desensitization is itself a risk โ it papers over the fact that this particular breach is structural, not operational.
Let me walk through the technical chain, because the details matter more than the dollar figure.
BIP39 seeds derive from 128 to 256 bits of entropy. Brute-forcing that space is computationally infeasible โ it would outlast the heat death of the universe. But in March 2021, an encoding error in Coldcard's firmware caused random number generation to silently switch to a weak fallback: entropy sourced from the device's serial number and internal clock. The keyspace collapsed from 2^128 to approximately 2^32 โ about four billion possibilities.
Four billion sounds large. To a modern computer, it is trivial. Block Research engineers estimated the damaged seed range and confirmed what the attacker already understood: enumerate the possibilities, convert each into a Bitcoin address, and scan chain data for balances. This is "address sweeping" at industrial scale. The attack cost almost nothing. The payoff was $70 million โ and the attacker is still holding roughly a quarter of it across four known addresses, in no apparent hurry to move.
Note: When entropy collapses by 96 orders of magnitude, "cold storage" becomes a public spreadsheet.
The forensic reconstruction by Galaxy Research and Block Research deserves attention. They rebuilt the attack timeline block by block. The sweeper wasn't broadcasting continuously โ three intermediate blocks showed no sweep activity, evidence that the attacker was deliberately batching transactions to avoid triggering exchange or surveillance alerts. This wasn't a reckless smash-and-grab. It was patient, structured extraction by someone who understood both the technical weakness and the operational economics of chain monitoring. Based on my experience auditing DeFi derivatives architectures in 2020, this silent-batching pattern is exactly what sophisticated operators do when they hold the better hand โ they throttle their own pace to stay under the monitoring threshold.
There is also a strong possibility the attacker pre-generated the entire vulnerable address space offline before broadcasting a single transaction. That would mean the sweep on July 30 was just execution โ the reconnaissance happened long before, unnoticed, on an air-gapped machine. The origin of this attack may extend back months.
Coinkite's response has been by-the-book: a fix firmware was released immediately. But here is the uncomfortable truth the industry doesn't want to confront. The fix only prevents newly created seeds from entering the compromised keyspace. The 1,196 swept wallets cannot be restored. And any Coldcard user who generated a seed between March 2021 and the patch may be exposed โ with no way to verify it at home. No test. No diagnostic. No practical method for an ordinary user to determine whether their seed lives in the compromised range.
That is the real story. Not the $70 million. The invisible exposure.
Any funds sent to a vulnerable address โ even transactions that arrive years from now โ remain at risk of immediate sweep. The 1,196 known victims are a floor, not a ceiling. If additional affected batches are identified, total losses could substantially exceed the current estimate. The deepest irony: most victims still can't confirm whether they're on the list.
Now the contrarian angle. The market wants to frame this as a Coldcard problem. It is not. It is a single-source RNG problem that spans the entire hardware wallet industry. Every vendor that relies on one internal entropy source carries the same structural risk. The only difference is that Coldcard got caught. When security depends on a single point of failure, the failure isn't a matter of "if" โ it's a matter of when, and how loudly.
CZ's public warning โ that even hardware wallets can have vulnerabilities โ drew predictable backlash from the Bitcoin maximalist community. But his underlying point deserves serious consideration: nothing is 100 percent. And that's not a dismissal of self-custody; it's a challenge to the mono-culture of security assumptions that has dominated Bitcoin storage for nearly a decade.
The BIP39 passphrase is the most revealing data point here. A strong passphrase would have protected even the compromised seeds โ without it, the attacker's derivation path hits a dead end. Yet the ecosystem has done a poor job of standardizing passphrase support across mobile wallets and applications. The single cheapest security upgrade available to Bitcoin users remains the most under-adopted โ and that's a systemic failure, not a user failure.
The structural winners haven't fully emerged. MPC wallets โ which split key material across multiple parties, eliminating the single-RNG vulnerability โ have the strongest narrative tailwind. Multi-signature platforms gain credibility. The "hardware wallet as absolute safekeeping" narrative has been replaced by something more honest: layered defense. Hardware plus passphrase. Multi-device. Multi-signature. Diversification of trust. Exchange custody may see short-term inflows from spooked self-custodians, which is deeply ironic given the 2022 failures that pushed users toward hardware wallets in the first place.
Coinkite's disclosure gap adds another layer of damage. The initial announcement failed to specify which models were affected, leaving Mk2 users in limbo. For a brand whose entire value proposition rests on radical transparency, that omission is self-inflicted damage.
Note: The next bull market will test whether hardware wallets have earned back the trust lost in this moment.
If you own a Coldcard and created a seed after March 2021, treat it as potentially compromised. Generate a new seed. Move the funds. Do not wait for a definitive affected-model list that may never come. For everyone else, the lesson is uncomfortable: self-custody is still superior to centralized custody, but "not your keys, not your coins" carries a hidden assumption โ that your keys are actually yours. Unpredictable. Private. Unreachable. When a single firmware bug reduces your keyspace to 32 bits, your keys stop being yours in any meaningful sense.
The $70 million was stolen in 41 minutes. The trust will take far longer to rebuild. And the next attack isn't coming through the same door โ it's already waiting in code that has been running undisturbed for years, quietly failing, until someone looks at it from the wrong angle. The question for every hardware wallet vendor is the same: what else has been silently broken since 2021? And for every user: are you sure you'd know?