Attackers drained approximately 1,367 BTC (worth roughly $88 to $89 million, as of August 2) from 4,585 Coldcard hardware wallet addresses between July 30 and August 2, 2026, according to Galaxy Research data cited by CoinDesk. The first wave lasted just 41 minutes and netted 1,082.65 BTC from 1,196 addresses. Galaxy Research identified approximately 600 suspected attacker-controlled addresses involved in the sweep. Canadian hardware wallet maker Coinkite issued patched firmware on July 31, but the damage was already done, and the flaw responsible had been sitting undetected in the codebase since a firmware update in March 2021.

The root cause was a single misconfiguration in the device's build system. When Coinkite integrated the libngu cryptography library, the production configuration set MICROPY_HW_ENABLE_RNG to zero, disabling the STM32 hardware random number generator. The library checked only whether the macro existed, not whether it was enabled. As a result, affected devices fell back silently to Yasmarang, a deterministic software algorithm bundled with MicroPython. Effective entropy on Mk3 devices collapsed to roughly 40 bits, far short of the 128-bit standard required for a secure 12-word BIP-39 seed phrase. Mk4, Mk5, and Q models fared better at around 72 bits, but still fell well below that threshold. A 40-bit keyspace can be exhausted in hours on standard consumer hardware.

The attack required no physical access to any device. According to Coinkite's security advisory, attackers pre-generated every possible seed from the weakened entropy pool, derived the corresponding Bitcoin addresses, cross-referenced those against the public blockchain, and then swept any funded wallets. It was entirely an offline operation against a predictable mathematical space. Affected firmware versions span Mk2 and Mk3 builds from 4.0.1 through 4.1.9, Mk4 and Mk5 builds below 5.6.0, and Q builds below 1.5.0Q.

Jameson Lopp of Bitcoin security firm Casa pointed to the nature of the flaw as the most troubling detail. The bug was not in the wallet's main code but in its build process, a layer that most external reviewers, including large language models used for automated code auditing, consistently overlooked. Lopp noted that critical vulnerabilities in the Coldcard build system went undetected for over a decade, a span that extends well beyond the five years specific to the RNG misconfiguration and points to a deeper, longer-standing blind spot in build-level security review. "An unfortunate aspect of being in the security business is that your 'paranoid' warnings tend to fall on deaf ears until catastrophe strikes," Lopp wrote. "Over a long enough time frame, bad things tend to happen. If your life savings are on the line, it's best to mitigate every edge case." Coinkite CEO NVK acknowledged that AI-assisted code review can now find latent bugs "at a speed outpacing even the industry's most seasoned experts," but he also noted that Coinkite's own internal AI review of the codebase failed to catch this flaw before the theft occurred. Upon confirming the vulnerability, NVK urged affected users: "Move your funds now...before reading further."

A logistical problem compounds the harm for existing users. Coinkite purges all customer purchase records after 120 days to limit its own exposure in the event of a data breach. That policy, which reflects reasonable privacy practice, means the company has no way to contact the buyers of vulnerable devices sold over the past five years. Users who purchased through resellers or informal channels are even harder to reach. Lopp flagged this directly: "Coinkite purges all their customer records after 120 days to protect against data breaches, which means they are unable to reach out to customers who bought vulnerable Coldcards over the past 5 years to warn them of this vulnerability." Updating firmware on a compromised device does not fix the problem; any seed generated under the faulty firmware must be abandoned entirely. Users need to generate a fresh seed on patched firmware and move all funds. One important exception applies: seeds created with 50 or more independent dice rolls, or protected by a strong BIP-39 passphrase, are not vulnerable to this specific flaw. Users who relied on either of those methods when setting up their devices do not need to migrate funds.

The incident is not without precedent. In 2023, the "Milk Sad" vulnerability exposed a near-identical class of failure: a weak pseudorandom number generator inside the Libbitcoin Explorer tool produced predictable seeds that attackers exploited to drain wallets at scale, using the same basic method of offline pre-computation against a constrained keyspace. The recurrence of this vulnerability class suggests that build-level entropy failures represent a systematic blind spot across the hardware wallet industry rather than a one-off anomaly.

The regional stakes are significant. India ranks first and Pakistan ranks third on the Chainalysis 2025 Global Crypto Adoption Index. Both countries have high rates of peer-to-peer and non-custodial usage, driven by currency volatility and either heavy taxation or regulatory friction. Coldcard has been recommended for technically sophisticated holders in these markets who wanted to avoid exchange counterparty risk. That risk calculus now looks different. Across Sub-Saharan Africa, where Nigeria alone has an estimated 33 million crypto users and several local exchanges have collapsed or faced banking restrictions in recent years, hardware wallets have long been positioned as the safer alternative to custodial platforms. Kenya, Ghana, and South Africa have similarly active Bitcoin communities with a strong preference for self-custody, meaning the Coldcard incident carries broad regional weight. Lorenzo Valente of ARK Invest suggested that "publicly-traded exchanges or ETFs may now be safer," but for most users in these regions, those options remain inaccessible or unaffordable. Multisig setups (requiring multiple independent devices to authorize a transaction) or software wallets seeded with externally verified entropy are more realistic near-term alternatives.

On-chain data adds a broader signal to the damage. CoinDesk's markets analysis noted a net increase in Bitcoin deposits to centralized exchanges in the days after July 30, reversing a self-custody trend that had built steadily since the FTX collapse in 2022. Whether that shift holds will depend on how the Bitcoin community processes the lesson at the center of this incident: open-source code is verifiable in principle, but the build system that compiles it can harbor invisible failures for years, surviving multiple audits and emerging only when an attacker with enough computing power and patience decides to look.