A firmware error introduced into Coldcard Bitcoin hardware wallets in March 2021 has enabled attackers to steal an estimated 1,816 BTC, worth roughly $115 to $130 million at Bitcoin prices that ranged from approximately $64,000 to $72,000 across the reporting period, across four attack waves between July 30 and August 3, 2026. The last of those waves showed signs of multiple competing actors rather than a single operator.

Galaxy Research issued a fresh warning on August 4 that total losses could climb further once Wave 4 is fully confirmed. The exploit required no physical access to victims' devices and no interaction from victims whatsoever.

How a Misconfigured Macro Broke Five Years of Security

The root cause traces to a single misconfiguration in Coldcard's production firmware. A macro called MICROPY_HW_ENABLE_RNG was set to zero, disabling the device's dedicated STM32 hardware random number generator.

The supporting code library (the libngu library), however, only checked whether the macro existed, not whether it was actually switched on.

As a result, wallet seed phrases were generated using MicroPython's software-based fallback, called Yasmarang, which was initialized from the chip's unique ID and timer registers. Both of those values are predictable.

The practical consequence was severe. Mk3 devices produced seeds with roughly 40 bits of actual randomness rather than the 128 bits required by the BIP-39 wallet standard (BIP-39 is the widely adopted specification for generating human-readable recovery phrases). Mk4, Mk5, and Q models fared better but still fell to around 72 bits, a reduction significant enough to make systematic enumeration feasible with sufficient computing resources.

Attackers generated all possible seeds offline, derived every associated Bitcoin address, cross-referenced those addresses against the publicly available blockchain, and swept any funded wallets. As one security researcher told CoinDesk: "Every step of that runs on the attacker's machine. The victim's device is not involved at any point and could be powered off in a safe on another continent."

Four Waves, Escalating Scale

The first wave hit at 1:10 UTC on July 30, draining roughly 594 BTC from around 500 addresses.

A second wave the same day, completing in 41 minutes and representing the most dramatic single sweep of the four, brought the cumulative total to 1,082 BTC across 1,196 addresses, equivalent to approximately $70.2 million. By August 2, that figure had climbed to approximately 1,367 BTC touching 4,585 addresses. Wave 4, on August 3, added another 449 BTC from 709 addresses.

On-chain analysts noted that Wave 4 showed signs of multiple competing actors rather than a single operator. Galaxy Research's Alex Thorn observed that "every victim in the confirmed set was routed to its own fresh destination address in a strict one-to-one mapping," a pattern distinct from earlier waves.

Transactions in Wave 4 also opted into replace-by-fee (RBF), a Bitcoin feature that allows a transaction to be replaced by a competing one with a higher fee before confirmation. That gave some alert victims a narrow window, measured in minutes, to broadcast a competing transaction and potentially reclaim funds.

The sweep rate in Wave 4 reached 13.8 transfers per block, up from a baseline of 0.3 transfers per block, roughly 45 times the normal rate, indicating fully automated execution.

As of reporting, proceeds from the first two waves remain unmoved in four on-chain addresses.

Coinkite's Response and What Users Must Do Now

Toronto-based Coinkite, the maker of Coldcard and one of the most respected names in Bitcoin self-custody hardware, shipped emergency firmware on July 31.

Patched versions are Mk3 firmware 4.2.0, Mk4 and Mk5 firmware 5.6.0, and Q firmware 1.5.0Q.

The company also destroyed remaining inventory of units manufactured with the vulnerable firmware and halted shipments.

CEO Rodolfo Novak (known as NVK) posted a blunt warning on July 31: "If you generated a seed using a Coldcard wallet, move your funds now, using our updated best practices, before reading further."

Critically, installing the patch does not fix a seed that was already generated under the flawed firmware. Any wallet created on affected firmware versions between March 2021 and July 2026 must be treated as permanently compromised.

Users must generate a fresh seed on patched firmware and transfer all funds to the new wallet. Affected firmware versions include Mk2 and Mk3 releases from 4.0.0 to 4.1.9, all Mk4 and Mk5 builds before 5.6.0, and all Q builds before 1.5.0Q. Entropy data specific to Mk2 devices has not been confirmed in available security advisories; Mk2 users should treat their devices as affected and migrate funds immediately.

A Harder Problem for South Asian and African Users

The implications extend well beyond Western markets. India ranks first on the Chainalysis 2025 Global Crypto Adoption Index, and Pakistan sits among the global leaders in peer-to-peer crypto volume. Across Nigeria, Kenya, and Ghana, hardware wallets are widely used not as a technical preference but as financial infrastructure, a practical alternative to unreliable banking systems, currency controls, and naira, shilling, or cedi volatility.

For users in those markets, the secondary problem is structural. The main alternative highlighted by analysts is a flight to regulated custodians or spot Bitcoin ETFs. David Lawrence, co-founder of Amicus, noted that incidents like this "push investors toward regulated products like BlackRock's iShares Bitcoin Trust rather than self-custody."

South Asian and African users generally cannot access US-listed ETFs and often lack affordable institutional custodians. That leaves a trust vacuum with no clean exit.

Guy Swann, a Bitcoin commentator, called this "the worst hit in bitcoin history to the most knowledgeable and 'properly secured' bitcoiners."

For emerging-market users who adopted self-custody precisely because institutional options failed them, that assessment carries particular weight.

What Comes Next

No attribution to state-sponsored actors has been established.

Forensic analysts identified that the attacker queried victim addresses using a paid account at a blockchain data provider, with provider logs matching the suspected workflow "with extraordinary specificity, down to the number, timing and sequence of requests," according to CoinDesk. A report by Bitcoin Magazine noted that AI tools may have been involved in systematically enumerating the reduced seed space at scale, though that claim remains unconfirmed.

Bitcoin's price has declined less than 1 percent since the attack began, suggesting markets have so far absorbed the immediate shock, but the broader fallout for hardware wallet trust is harder to quantify. For context, security researchers tracked 207 separate incidents in the first half of 2026, the highest half-year count on record, though total losses of $972 million were less than half those recorded in the same period of 2025 ($2.3 billion).

Coinkite's statement acknowledged the scale of the damage plainly: "The last three days have been some of the hardest in this company's history, and for a lot of the people reading this, they've been something much worse." Whether the self-custody model can recover confidence among the users who depended on it most is the question that will define the next chapter for Bitcoin's retail security ecosystem.