The $116 Million Coldcard Hack: How a Firmware Bug Collapsed Bitcoin Wallet Entropy

Instead of cryptographically secure entropy, the software fallback derived seeds from predictable inputs — the chip's serial number and internal clock . This collapsed the effective entropy from the intended 256 bits down to a tiny, enumerable search space. A seed phrase generated on an affected device was no longer effectively random . Because the serial number and clock values could be guessed or brute-forced offline, an attacker could reconstruct the full seed phrase (12 or 24 words) for any vulnerable wallet without ever touching the physical device .

How the Attack Worked

The attacker(s) performed large-scale offline seed enumeration — computing every possible weak seed for the vulnerable firmware range, deriving the corresponding private keys, scanning for balances, and sweeping the funds. On-chain analytics firm Galaxy Research mapped the primary thefts and identified at least three distinct attacker clusters .

Multiple sources reported that AI-assisted code auditing may have been used to discover the flaw, and AI tools almost certainly enabled the massive offline seed enumeration and balance scanning that made the scale of the theft feasible . The AI connection has sparked debate about a new era of automated vulnerability discovery.

Who Was Affected

• Affected devices: Coldcard Mk2 and Mk3 hardware wallets running firmware versions 4.0.1 (March 2021) through 4.1.9, and up to 5.0.3 per the broader advisory .
• Safe users: Coldcard users who generated their seeds using dice rolls (custom entropy), added a BIP 39 passphrase, or used firmware 4.2.0 or later were not vulnerable. Mk4 devices were never affected .
• Scale: Over 5,200 individual Bitcoin addresses across 4,500+ wallets were drained .
• The largest single sweep occurred in the early hours of July 30, when 1,196 addresses were emptied of 1,082.65 BTC (~$70.2M) in just 41 minutes .

The Response

Coinkite's response was swift. It issued a public security advisory on July 30, 2026, warning all Mk2/Mk3 users that funds in wallets created on vulnerable firmware versions were at risk . The company urged affected users to immediately move funds to new wallets generated on fixed firmware or with dice-roll entropy , released patched firmware (version 5.0.4+) to restore proper hardware TRNG usage , and published a full timeline tracing the bug to the March 2021 commit . It also engaged with blockchain analytics firms including Galaxy Research and Block to track the stolen funds .

Victim response was chaotic. Many users were unaware of the vulnerability until their funds were already drained. Social media saw widespread reports of empty wallets on the afternoon of July 30 . Multiple competing check-tool websites appeared, some of which were themselves phishing attempts. Little recourse exists for victims of a seed-reconstruction attack since no centralized custodian holds the funds.

Ledger (competitor response): Ledger publicly stated its devices were not affected, noting that Ledger uses a certified TRNG built into a secure element chip generating full 256-bit entropy for every recovery phrase .

Broader Market and Self-Custody Impact

• Bitcoin price: The hack did not cause a major BTC price crash, though it added to general bearish sentiment in an already volatile period .
• Hardware wallet industry: The incident dealt a heavy blow to Coldcard's reputation as the gold standard for "paranoid-level" Bitcoin self-custody . It triggered broader questions about whether air-gapped hardware wallets are truly safe if firmware bugs can silently destroy entropy.
• AI security debate: The exploit is widely discussed as a watershed moment — AI-assisted vulnerability discovery is now seen as a realistic threat to any open-source hardware wallet firmware .
• Self-custody lessons: Industry voices have renewed calls for multi-signature setups, passphrase-protected seeds, and verified hardware entropy sources as minimum best practices .

Current Status of the Stolen Funds (as of August 7, 2026)

The stolen funds remain overwhelmingly unspent, monitored, and effectively frozen in plain sight, despite some recent laundering activity.

• Largest cluster: The primary attacker controls 1,159 BTC ($72.7M) across seven addresses. As of August 5, none of that BTC had entered any mixer or been cashed out through an exchange .
• Second cluster (64 BTC): Another attacker group began moving funds on August 4–6, routing 64 BTC through Wasabi mixer and 200 ETH through Tornado Cash . This suggests smaller-scale operators who obtained partial seed databases are attempting to launder first.
• Third cluster (30 BTC): On August 7, an attacker believed to hold ~2,055 BTC moved 30.185 BTC (~$1.94M) to a new wallet — the first movement in weeks from that cluster .
• Overall: As of August 4, approximately 90% of the roughly 1,600 BTC stolen had not moved since the initial sweep, 0 BTC had been cashed out through any known exchange, and 0 BTC had entered a mixer from the largest cluster . The bulk of the funds remain in monitored addresses under constant blockchain surveillance by Galaxy Research, Block, and law enforcement .

Because the funds are on public Bitcoin addresses that are being tracked in real time, the attackers face enormous difficulty cashing out significant amounts without detection. The funds are effectively frozen — but they have not been returned to victims, and no law enforcement seizure has been publicly confirmed as of the latest reporting.