A vulnerability in Coldcard hardware wallets has led to the theft of approximately 1,367 BTC, worth around $89 million, from over 4,500 addresses since coordinated attacks began on July 30. The bug, quietly sitting in firmware since March 2021, weakened the randomness used to generate wallet seeds so severely that attackers were eventually able to crack them.

Foundation Devices CEO Zach Herbert pointed to the open-source nature of Bitcoin wallet code as the reason the community was able to detect the issue and respond as quickly as it did. His argument: proprietary firmware would have made the problem harder to find and even harder to fix.

What went wrong inside Coldcard’s firmware

When you set up a hardware wallet, the device generates a “seed,” which is essentially the master key to all your Bitcoin. That seed needs to be random. The industry standard is 128 bits of entropy, which in practical terms means the number of possible seeds is so astronomically large that brute-forcing your way to the right one would take longer than the age of the universe.

Coldcard’s firmware version 4.0.1, released in March 2021 by Canadian manufacturer Coinkite, introduced a bug during a significant code rewrite. Instead of generating seeds with 128 bits of entropy, affected devices, specifically the Mk2 and Mk3 models, produced seeds with roughly 72 bits.

The difference between 128 bits and 72 bits of entropy sounds minor if you’re not a cryptographer. It is not. Each missing bit cuts the number of possibilities in half. Losing 56 bits means the search space shrank by a factor of roughly 72 quadrillion. That’s the kind of math that turns a theoretical vulnerability into a practical one.

The attack and the response

Coordinated attacks exploiting the weakness began on July 30, 2026. Within days, approximately 1,367 BTC had been siphoned from wallets whose seeds were generated on affected firmware versions. At the time, that haul was valued at around $89 million.

The speed of the drain across more than 4,500 separate addresses suggests the attackers had been preparing for some time, likely pre-computing vulnerable seeds before executing the thefts in rapid succession.

Coinkite released patched firmware on July 31, just one day after the attacks became apparent. The fixes arrived as version 5.6.0 for Mk4 and Mk5 devices, version 1.5.0Q for the Q model, and version 4.2.0 for the older Mk3. But updating firmware alone doesn’t fix the problem. Users who generated seeds on the vulnerable firmware still need to migrate to entirely new seeds, because the old ones remain compromised regardless of what software the device is now running.

Seeds generated using at least 50 independent fair dice rolls were not affected. Neither were seeds created on firmware versions prior to 4.0.1. The vulnerability was specific to the entropy generation code introduced in the 2021 rewrite.

The open-source argument

Zach Herbert, who runs Foundation Devices, a competing hardware wallet manufacturer, weighed in on the incident by emphasizing that open-source code was central to the community’s ability to respond quickly. His position is straightforward: when wallet firmware is publicly auditable, vulnerabilities get found faster, patches get scrutinized before deployment, and the entire ecosystem benefits from collective oversight.

Open source code helps the Bitcoin community respond quickly and may prevent future security incidents.

Foundation Devices makes the Passport wallet, a direct competitor to Coldcard. This vulnerability sat in production firmware for over five years before it was exploited. The rapid patch turnaround — one day from attack to fix — supports the idea that transparent codebases accelerate incident response.

What this means for hardware wallet users

For users still using Coldcard devices with seeds generated after March 2021, the action items are clear: update firmware immediately and, more importantly, generate entirely new seeds. Simply patching the firmware without migrating to a new seed leaves the underlying vulnerability intact. Users who employed at least 50 independent fair dice rolls for their seed generation were never dependent on the device’s internal random number generator and are not affected.