Coldcard Firmware Bug Behind $70M Bitcoin Theft

On July 30, an attacker drained 1,196 Bitcoin addresses in 41 minutes, taking 1,082.65 BTC — approximately $70.2 million at the time. Galaxy Research mapped the sweep and traced it to a firmware integration error in the Coldcard hardware wallet introduced in March 2021.

The flaw is conceptually simple in the way that the most consequential ones tend to be. Coldcard, the Bitcoin-only hardware wallet built by Canadian firm Coinkite, is designed around a core assumption: private key generation draws entropy from a hardware random number generator on the device itself. The March 2021 change routed that seed generation to a deterministic software pseudorandom number generator instead. The resulting keys appeared valid. They passed every visible check. The entropy they were built on could be reconstructed offline by anyone who understood the PRNG’s state space.

The mechanics of the sweep

The 41-minute drain window is the operational signature of bulk key reconstruction, not a targeted attack on individual wallets. When an attacker can derive private keys from a known, reproducible algorithm rather than cracking each key individually, the limiting factor becomes bandwidth and blockchain throughput, not cryptographic work. You enumerate the affected PRNG outputs, derive the private keys, identify which addresses hold funds, and sweep them in sequence. Forty-one minutes for 1,196 addresses fits that model.

Galaxy Research documented the address-drain sequence. The Hacker News has the primary coverage; Coinkite’s advisory is the authoritative source on which firmware versions are affected and what remediation steps apply — that’s where affected users should start.

What the trust model actually requires

Hardware wallets exist because keeping private key material off a general-purpose computer matters. The threat model they address is real. What this incident illustrates is that the security guarantee runs precisely as deep as the entropy used to generate the keys, and no further.

A 2021 firmware error that routed seed generation to a software PRNG bypassed the foundation of that model without breaking anything visible at the surface. The wallet still functioned. The keys were still valid. The only thing missing was the actual unpredictability the hardware RNG was supposed to provide, and that absence didn’t manifest as an error — it manifested as a sweep four-plus years later.

This is the same mistake, different decade, different substrate. It is not a hardware wallet problem specifically; it is a problem of entropy paths that get quietly rerouted and then produce valid-looking output for years. Reviewing the integrity of those paths does not fit neatly into a standard audit checklist, which is part of why it keeps happening.

For Coldcard users

Coinkite is the authoritative source on affected firmware versions and remediation. If a device ran firmware released in or after March 2021, users need to determine whether their key generation used the affected code path. If it did, funds in affected addresses need to move — to keys generated on unaffected firmware, or to a different device entirely. The PRNG state for those keys is, by the nature of this vulnerability, potentially reconstructible. Waiting is not a mitigation.

Whether July 30 represents the first exploitation of this flaw or the visible tail of a longer window is, as of publication, unconfirmed.