The 2026 Coldcard Breach: How a 2021 Firmware Bug Led to Bitcoin's Worst Self-Custody Failure

On July 30, 2026, an attacker began draining Bitcoin from Coldcard hardware wallets — devices marketed as one of the most secure ways to self-custody cryptocurrency. Over the following days, four waves of thefts hit more than 7,300 addresses, with total losses reaching an estimated 2,055 BTC, worth roughly $130 million at prevailing prices . The root cause was a single firmware build error introduced five years earlier, in March 2021, that silently crippled the randomness of seed-phrase generation .

The Firmware Vulnerability: A Broken Entropy Chain

A March 2021 firmware integration error caused Coldcard devices to bypass their dedicated hardware entropy chip during seed-phrase generation and fall back to a software-based pseudo-random number generator (PRNG) instead . The bug affected firmware versions 4.0.1 through 4.1.9, with later analysis extending the scope to versions up to 5.0.3 on Mk3 units .

On affected Mk3 devices, effective entropy collapsed from the expected 128 bits to as low as ~40 bits, making seed phrases brute-forceable offline without any physical access, phishing, or malware . Coinkite's own analysis later confirmed that some Mk3 seeds had only about 40 bits of entropy . A firmware check meant to verify the hardware RNG was active was itself broken by the same update, allowing the bug to go undetected for over five years .

Technical Specifics

Block (Square) researchers published a detailed technical analysis on July 31, tracing the root cause to an integration error in Coldcard's random number generation code. "COLDCARD firmware contains an RNG integration error that causes ngu.random to use MicroPython's deterministic Yasmarang fallback instead of the STM32 hardware RNG," Block's report explained . This meant seed phrases were generated using predictable values — including the microcontroller's serial number and system counter — rather than true hardware randomness .

How Much Was Stolen and Why Estimates Vary

Loss estimates range from $38 million to $130 million, and the wide spread is not contradictory — it reflects different snapshots as the attack unfolded and researchers expanded their on-chain analysis.

Estimate (BTC)	Approx. USD	Source / Timing
594 BTC	~$38M	Coinkite's first advisory, 500+ dormant wallets
1,082.65 BTC	~$70.2M	Galaxy Research, first 41-minute major drain
1,367 BTC	~$88–89M	Galaxy Research, early August tracking
1,816 BTC	~$116M	Forbes / TRM Labs, 5,200+ addresses
2,055 BTC	~$130M	Galaxy Research, 4 waves, 7,300+ addresses

Why the Numbers Differ

• Waves over time: The attacker(s) executed at least four distinct sweeps between July 30 and early August, with each new wave expanding the known total .
• BTC price volatility: Reports denominate losses at the Bitcoin price prevailing at the time of reporting, which fluctuated during the incident period.
• Gradual address discovery: Researchers identified vulnerable addresses incrementally. Early counts included only obviously drained wallets, while later on-chain analysis by Galaxy Research and TRM Labs uncovered additional affected seeds that weren't immediately exploited .
• Coinkite's data gap: Coinkite stated it lacks the customer data needed to independently verify total losses due to its privacy-first design, so it has declined to confirm any specific figure .

How Coinkite Responded

Coinkite's response was rapid, though the company faced a fundamental limitation: because the vulnerability was in seed generation itself, a firmware patch could not fix seeds that had already been created .

• July 30, 2026: Coinkite issued an emergency security advisory for Coldcard Mk3 owners, warning that funds tied to affected firmware versions were at risk .
• July 31, 2026: Coinkite released patched firmware for all affected models (4.2.0+ for Mk3; 5.6.0+ for Mk4 and Mk5; 1.5.0Q+ for Q) and published a technical backgrounder detailing the root cause . The company also posted an open letter calling the preceding days "some of the hardest in this company's history" and urged all users who had generated seeds on Coldcard devices to immediately migrate funds to newly generated seeds .
• Early August: Coinkite reversed its long-standing data-deletion policy, announcing it would now retain user data to meet potential legal obligations arising from the breach .
• Ongoing: Coinkite committed to publishing a full technical post-mortem once it is safe to do so, but had not yet released it as of mid-August 2026 .

How Researchers Responded

The broader security research community mobilized quickly to map the scale of the exploit and trace its root cause.

• Galaxy Research led on-chain forensics, mapping the initial 41-minute drain of 1,082.65 BTC and progressively expanding the known scope across four attack waves .
• Block (Square) researchers published a detailed technical analysis on July 31 tracing the root cause to the RNG integration error .
• TRM Labs published a comprehensive incident analysis estimating ~1,816 BTC stolen from over 5,200 addresses, calling it the largest hardware wallet exploit on record .
• A community-run incident record was established to track the evolving scope and technical details .

Broader Industry Fallout

• Bitcoin.org removed Coldcard and Coldcard Q from its official wallet listings after the incident, citing its published wallet-listing criteria .
• The hack reignited the self-custody debate, with some analysts arguing it may push investors toward institutional custody solutions or Bitcoin ETFs . Bitcoin new-wallet creation nevertheless spiked to a 2026 high in the aftermath, driven partly by victims migrating to new seed phrases .
• Cybersecurity stocks rallied on expectations of increased hardware security scrutiny, while BTC and ETH prices remained broadly stable .
• Industry experts and some media outlets speculated that AI-assisted brute-forcing may have been used to accelerate the seed cracking, though no definitive proof has been published . Coinkite itself acknowledged that even an internal review using "one of the best available AI models" to scan firmware code had failed to detect the flaw or any other serious vulnerability .

Key Lessons for Hardware Wallet Users

• A firmware update does not fix an already-compromised seed. Coinkite was unambiguous: "Updating the firmware does not repair a seed that was generated by affected firmware. A new seed must be generated on the latest firmware, and funds moved to that new wallet" .
• Manual entropy sources (dice rolls, BIP 39 passphrases) likely mitigated the vulnerability. Users who generated their own entropy via Coldcard's dice-roll feature or used a BIP 39 extra passphrase were significantly less exposed .
• Privacy-first design has costs. Coinkite's inability to verify loss totals or contact affected customers stemmed from its design principle of not retaining user data — a tradeoff that became a liability in a crisis .