A five-year-old firmware defect in Coinkite's Coldcard hardware wallets has resulted in the theft of approximately 1,816 BTC — valued at roughly $116 million — from more than 5,200 addresses across four attack waves beginning July 30, 2026. According to TRM Labs, the incident ranks as the third-l...
"This is the worst hit in bitcoin history to the most knowledgeable and 'properly secured' bitcoiners. This isn't an exchange getting hacked because of hot keys. This is thousands of individuals having their personal private keys recreated out from underneath them." — Guy Swann, Bitcoin Commentator
A five-year-old firmware defect in Coinkite's Coldcard hardware wallets has resulted in the theft of approximately 1,816 BTC — valued at roughly $116 million — from more than 5,200 addresses across four attack waves beginning July 30, 2026. According to TRM Labs, the incident ranks as the third-largest crypto hack of 2026 and the largest hardware wallet exploit on record.
The vulnerability, introduced in a March 2021 commit, caused affected devices to bypass their dedicated hardware random number generator during seed phrase creation, instead falling back on a predictable software substitute. The result was a collapse in effective key strength from a designed 128 bits to as few as 40 bits — low enough to brute-force without physical access to the device.
The fallout has reignited a long-dormant debate about Bitcoin custody models. U.S.-listed Bitcoin ETFs posted $850 million in weekly inflows in the hack's aftermath, according to Bloomberg, while on-chain data showed 11,163 BTC in net exchange inflows on July 31. The incident exposes a structural tension: self-custody eliminates counterparty risk but introduces firmware, supply-chain, and operational risks that most retail holders lack the capacity to audit.
The root cause traces to a single code commit dated March 1, 2021, in the libngu cryptographic library used by Coldcard devices. The commit replaced the hardware random number generator (HRNG) — a dedicated silicon component designed to produce cryptographically secure entropy — with a software pseudorandom number generator (PRNG) seeded from hardcoded constants.
According to TRM Labs' technical analysis, the build configuration error affected firmware version 4.0.0 and subsequent releases across Coldcard Mk3, Mk4, Mk5, and Q product lines. The practical effect: seed phrases generated on these devices during the vulnerability window (March 2021 to July 31, 2026, when patched firmware was released) contained insufficient randomness to resist brute-force attacks.
The designed entropy of a BIP-39 seed phrase is 128 bits, which requires approximately 3.4 × 10^38 attempts to brute-force. With the software PRNG and hardcoded seed values, effective entropy dropped to approximately 40 bits on older devices — roughly 1.1 trillion possibilities. Modern computing clusters can exhaust that search space in hours.
The attack required no physical access to victim devices. Attackers replicated the same flawed PRNG process on their own hardware, generated candidate seed phrases, derived the corresponding public keys, and checked them against the Bitcoin blockchain for funded addresses.
TRM Labs documented four distinct attack waves:
| Wave | Date | BTC Drained | Addresses Hit | Duration | |------|------|-------------|---------------|----------| | 1 | July 30 | ~594 BTC (~$38M) | ~500 | ~25 minutes | | 2 | July 31 | ~488 BTC (~$31M) | ~1,700 | ~41 minutes | | 3 | August 1 | ~412 BTC (~$26M) | ~1,800 | ~3 hours | | 4 | August 2-3 | ~322 BTC (~$21M) | ~1,200 | ~6 hours |
Total estimated losses: ~1,816 BTC (~$116 million) from 5,200+ addresses.
Transaction patterns across the four waves differ in construction methodology, leading TRM Labs to conclude that multiple independent attackers may be involved. The firm has not attributed the theft to a specific actor. Galaxy Research separately identified at least 15 distinct attacker entities exploiting the vulnerability.
Laundering activity has been limited. As of August 10, TRM tracked a single 64.9 BTC deposit into Wasabi Wallet's CoinJoin mixer and 200 ETH deposited to Tornado Cash, both on August 4. The majority of stolen funds remain consolidated in a small number of attacker-controlled addresses with minimal onward movement.
On-chain analyst Willy Woo estimated recovery odds at 20% to 40%, warning that any restitution process would likely take years.
On August 4, Bitcoin developer James O'Beirne published a forensic analysis of the libngu repository's commit history. O'Beirne presented evidence that the pseudonymous GitHub account "Switck," which authored the March 2021 commit introducing the vulnerability, shares a GPG signing key with Coinkite CTO and co-founder Peter Gray.
Specifically, 58 commits attributed to "Switck" contain valid cryptographic signatures produced by Gray's personal GPG key, according to O'Beirne's analysis. O'Beirne further stated that he had discovered and reported the vulnerability to Gray during a firmware audit in 2025, but the report was dismissed.
Coinkite has not publicly addressed the code attribution claims. CEO Rodolfo Novak issued a public apology on X (formerly Twitter), stating the company was "heartbroken" and taking "full accountability for the firmware bug." On July 31, Novak advised all users who generated seeds on Coldcard devices to migrate funds immediately. As of August 6, Coinkite declined to estimate total customer losses, stating through Bloomberg that it was "heads down helping affected customers."
The company suspended its automatic customer-data deletion policy to preserve records for anticipated litigation.
The hack triggered measurable shifts in Bitcoin custody patterns.
ETF Inflows: U.S.-listed Bitcoin exchange-traded funds attracted $850 million in weekly net inflows in the days following the exploit, according to Bloomberg — the strongest weekly intake since April 2026. Funds managed by BlackRock, Fidelity, Grayscale, and Morgan Stanley collectively received $626 million of that total.
Exchange Inflows: Net bitcoin inflows to centralized exchanges totaled 11,163 BTC on July 31, the day after the first attack wave, according to on-chain tracking by OKX. However, daily exchange inflow averages over the subsequent week ($1.55 billion) remained slightly below July's average of $1.67 billion, suggesting repositioning rather than panic selling.
Network Activity: Bitcoin's active address count rose to 751,000, with 2.27 million new wallets created in the week following the exploit — the strongest network activity reading of 2026, according to Bitcoin.com. This indicates broad migration to fresh key material rather than abandonment of the network.
Bitcoin ETF flows on August 10 reversed to -$144.6 million, breaking a 7-day positive streak. This suggests the custody-fear premium was short-lived.
The exploit has split the Bitcoin community along a familiar fault line, though with new empirical weight on both sides.
Pro-Institutional Custody: Former Binance CEO Changpeng Zhao (CZ) wrote that "it is statistically safer to store coins on exchanges than to self-custody," though he caveated that the comparison depends on data accuracy. ARK Invest Director Lorenzo Valente argued that "the self-custodial hardware space is a disaster at this point," noting that consumers have "traded counterparty risk for software risk, hardware risk, supply-chain risk, phishing risk, backup risk, and the possibility of losing everything through one mistake."
David Lawrence, co-founder of Amicus, characterized the incident as "hugely damaging to the people who believe that 8 billion people will hold their Bitcoin in cold storage in the future." CoinDesk reported that analysts see the exploit boosting demand for regulated Bitcoin exposure products.
Pro-Self-Custody: Willy Woo countered that "custodial solutions have their place" but that self-custody remains the "only path to genuinely sovereign Bitcoin ownership." He argued the incident reflected a vendor failure, not a protocol failure. Taproot developer Udi Wertheimer offered a more measured take: "The idea of your bitcoin resting easy in some secret location while you enjoy life not worrying about it is currently unrealistic."
Middle Ground: Onramp CEO Michael Tanguma stated that "self-custody and ETFs share the same flaw: a single point of failure." He advocated for multi-institutional, multi-signature custody arrangements — a model where three separate institutions each hold one key, requiring two of three to authorize transactions. This distributes trust across entities without concentrating it in a single hardware device or custodian.
The hardware wallet market was valued at $720 million in 2026, according to Mordor Intelligence, growing at a projected 29.05% CAGR through 2031. Ledger, the market leader, has sold over 8 million devices across 165 countries.
The broader self-custody adoption gap is significant. According to Ledger's own estimates, of 400 million global crypto users, only 30 million practice self-custody, and a mere 10 million do so with hardware wallets or comparable security measures. The individual user segment accounts for 73% of hardware wallet purchases.
The Coldcard exploit affects an undetermined fraction of devices sold between March 2021 and July 2026. Coinkite has not disclosed total units sold during this period. Coldcard has historically positioned itself as a Bitcoin-only, security-maximalist device — making its user base disproportionately composed of high-conviction, long-term Bitcoin holders.
Coinkite faces the prospect of class-action litigation. The company suspended automatic customer-data deletion on August 7 to preserve records, according to CryptoTimes. Legal experts remain divided on liability. Some argue Coinkite could face negligence claims given the five-year window between the bug's introduction and its detection; others question whether hardware wallet makers fall under existing product liability frameworks.
The incident may also accelerate regulatory attention. The $720 million hardware wallet market currently operates with minimal oversight. No jurisdiction requires hardware wallet manufacturers to submit firmware to independent security audits, disclose vulnerability windows, or maintain insurance pools for exploit losses. A Coinbase Custody user who loses 10 BTC has a claim against the custodian's insurance; a Coldcard user who loses 10 BTC has none.
From an economic-value perspective, the exploit exposes an unpriced externality in the self-custody stack. Users pay $150-$250 for a hardware wallet but bear unbounded downside risk from firmware failures they cannot audit. The value proposition of hardware wallets — eliminating counterparty risk — depends on the assumption that firmware integrity risk is negligible. The Coldcard incident demonstrates that assumption is falsifiable.
The Coldcard exploit is not evidence that self-custody is categorically inferior to institutional custody. It is evidence that the self-custody security model carries a class of risk — firmware integrity, supply-chain assurance, entropy quality — that most retail holders cannot independently verify. The FTX collapse demonstrated the cost of counterparty risk. The Coldcard hack demonstrates the cost of vendor-trust risk. Both are real. Neither is zero.
The $850 million in ETF inflows following the hack suggests the marginal Bitcoin buyer is choosing regulated wrappers over direct key management. Whether this represents rational risk assessment or a reflexive flight from complexity remains an open question. What the data shows: when the self-custody stack fails, capital moves to intermediaries. The economic implications of that preference — for Bitcoin's censorship resistance, for the distribution of custody infrastructure, for the regulatory surface area of the asset class — will take years to resolve.
For the 5,200+ affected users, the immediate question is simpler and more painful: who, if anyone, pays for a firmware bug that sat in production for five years.