A dormant RNG in Coldcard hardware wallets left users exposed, and the losses prove it

Ryan Sean Adams raised an alarm about Coldcard hardware wallets that, at the time, might have sounded like a technical footnote. The devices, he noted, shipped with a random number generator that was meaningfully better than the fallback option, but that superior generator was never activated. The implication was straightforward and serious: users who believed they were protected by high-grade hardware entropy were not.

That claim has since found substantial corroboration in the public record. According to Block’s engineering blog, a firmware misconfiguration in Coldcard devices caused the hardware’s stronger RNG to remain effectively dormant. Rather than drawing on the better source of randomness the device contained, the firmware fell back to a weak pseudorandom number generator. The difference between a strong hardware RNG and a weak software fallback is not a matter of degree. It is the difference between a private key that is computationally infeasible to reproduce and one that an attacker, given enough knowledge of the fallback’s behavior, can reconstruct.

The consequences were not theoretical. Coverage from CoinDesk, Decrypt, and CryptoBriefing, alongside GitHub commit records, points to approximately $70 million in losses drawn from 1,196 wallets, with the bulk of the damage occurring in July 2026. These were not users who cut corners on security. Coldcard is a product marketed specifically to people who take self-custody seriously. The hardware wallet is the choice of users who distrust software solutions and want the strongest available isolation between their private keys and any networked device. That positioning made the dormant RNG more damaging, not less: the product’s reputation attracted precisely the users who would assume the hardware entropy was functioning as designed.

It had a um a random number generator that was like much better, but it was not actually switched on. Ryan Sean Adams

The flaw itself sits at a frustrating intersection of hardware capability and firmware execution. The superior RNG existed. It was present in the device. The engineering decision, or the oversight, that left it unactivated is not yet fully detailed in public disclosures, but the effect is documented clearly enough. When a device’s randomness source is weak and predictable, the private keys it generates inherit that weakness. An attacker who understands the statistical properties of the fallback generator can work through the space of probable outputs far faster than a brute-force search against a truly random key would allow. Across 1,196 wallets, that narrowed search space translated to real losses at scale.

Adams’ framing deserves credit for identifying the core structural problem: a better component was present but disabled. That is a different category of failure from a design that never incorporated strong entropy in the first place. It suggests the device passed some internal bar for having the right hardware, then failed at the separate step of ensuring that hardware was actually used. Whether the cause was a configuration error, a testing gap, or something else in the firmware development process, the outcome was the same for the users affected.

What the Coldcard episode clarifies, more than any abstract debate about hardware wallet security, is that the presence of a security feature and its activation are not the same thing. Users and reviewers who inspect hardware specifications for the presence of a quality RNG are not necessarily checking whether that RNG is the one the firmware calls when it generates keys. That gap, between what a device contains and what it uses, is precisely where this failure lived. The Block engineering blog’s documentation of the fallback behavior and the GitHub commit history together provide enough detail for the security community to understand the mechanism. The $70 million figure provides the scale.

The broader takeaway for the self-custody ecosystem is uncomfortable but necessary. Hardware wallets derive much of their value from the assumption that their critical operations are both isolated and correctly implemented. Isolation without correct implementation offers a narrower margin of safety than users tend to assume. Auditing a device’s firmware for the correctness of its entropy calls, not just for the presence of capable hardware, is work the security community will need to treat as standard practice rather than an edge case. Adams identified the shape of the problem before the losses made it undeniable.