Coldcard Bitcoin Wallet Flaw Enabled $70 Million Drain Without Touching Devices

More than 1,000 bitcoin—about $70 million—was siphoned from 1,196 wallets in a 41-minute burst on July 30, nearly doubling initial estimates. Galaxy Research reconstructed the incident on Friday, reporting that 1,082.65 BTC was swept between 01:10 and 01:51 UTC across six blocks. Three blocks in the middle contained no related activity, a pattern that points to transactions being broadcast in batches. The stolen funds remain parked across four addresses and have not moved; earlier coverage tracked only one of those, which is why the total later rose. While the dollar value is smaller than some of 2026's biggest crypto thefts, investigators say the method is what makes this case stand out. Typical crypto heists require access—breaching an exchange, exploiting a smart contract, or phishing a private key. Hardware wallets sell the opposite: keys kept on a device that never connects to the internet, leaving an attacker nothing to touch. In this episode, the wallet hardware was never contacted. Instead, the weakness stemmed from how some Coldcard devices generated their seed—the master secret from which all addresses and private keys are deterministically derived. Coldcard firmware was designed to source that seed from a dedicated hardware random number generator. An internal build setting instructed the firmware to bypass that generator, and a check in a supporting library validated only that the setting existed, not that it was enabled. As a result, key generation fell back to a basic software substitute seeded by the chip's serial number and clock registers. That substitution proved costly. The serial number is fixed factory metadata, and clock values are timing states an attacker can narrow down or reproduce on their own device. The outcome was a dramatic collapse in the possible keyspace—from effectively unbounded to finite and searchable. Security teams determined that seeds could be pinned down on older Mk2 and Mk3 models, while for the Mk4, Q and Mk5 they estimated a range of roughly four billion possibilities. Four billion is large by human intuition, but tractable for modern compute. With the attack fully off-device, an operator can generate candidate seeds on their own hardware, derive the corresponding addresses, and compare them against the public blockchain. Galaxy's analysis suggests this is what happened. Of the drained wallets, 1,183 used the native SegWit address format, seven used an older standard, and six used an even older one. Investigators say hitting multiple address formats at once is consistent with systematic enumeration—testing each candidate seed across every derivation path it might produce—rather than targeting a specific victim. Galaxy warned additional waves are likely unless affected users move funds. Owners also can't reliably test whether they're exposed: there is no check that can confirm whether a given wallet's seed falls inside the reproducible range. Coinkite, Coldcard's manufacturer, has warned Mk3 owners and says newer devices are unaffected. Block's report, though, places the Mk2, Mk4, Q and Mk5 within scope as well. Until the discrepancy is resolved, users who generated seeds on impacted firmware are being told to assume exposure rather than attempt to prove safety. Investigators say the attacker may have left a trail. Block's Clay Garrett wrote on X that the operator used a paid account at a "well known blockchain data provider" to query source addresses during the sweeps. According to Garrett, internal logs at the provider matched the suspected workflow with "extraordinary specificity," including the number, timing and sequence of requests. The provider appears to have been responding to normal-looking queries with no obvious indicator of criminal intent. Block said it has shared the information with authorities. The incident is fueling a broader reassessment of cold storage assumptions. Cold wallets promise private keys are unguessable, but many users have treated that as a guarantee keys are unreachable. As the cost of finding and exploiting weaknesses in key generation falls, keeping a key offline may be the easier half of security. Anthropic published research Tuesday showing one of its models halved the security of a candidate post-quantum algorithm in 60 hours—against a design that had survived two years of expert review.