Canadian bitcoin holders have suffered 25% of all attributable losses in an ongoing exploit affecting certain Coldcard hardware wallets, making them the largest national group impacted by the incident.
The concentration of losses corresponds with the strong presence of Coldcard parent company Coinkite in Canada. The company is headquartered in Toronto.
Chainalysis tracking shows Australia is the second-worst affected country, accounting for between 15% and 20% of the total damage. The United States and Thailand follow, with each responsible for losses estimated at between 10% and 15%.
Although the breach has hit English-speaking countries and other early-adopting bitcoin markets particularly hard, losses have been recorded across the world. Western Europe, Latin America and major African cryptocurrency centres, including Nigeria and South Africa, have also been affected.
The total value of assets stolen in the incident has now reached $116m. Galaxy Research’s analysis identified a firmware update released in March 2021 as the source of the weakness. The update introduced a new random number generator which, because of an implementation error, became the single point of failure used in the attack.
“The problem was it was wired incorrectly and defaulted to a weaker one instead. It failed silently with no warning. Nobody knew their private keys were being generated with low entropy,” Galaxy Research stated. “Five years later, an attacker swept $70M from 1200 wallets in 41 minutes.”
Natalie Newson, senior blockchain investigator at CertiK, said the flaw went undetected for more than five years because of a configuration mistake involving the setting MICROPY_HW_ENABLE_RNG, which was set to zero.
“To a static guard checking #ifndef, a macro set to 0 is still defined,” Newson explained. “The safety check evaluated to true, suppressing the #error guard and allowing the build system to proceed as if everything was properly configured.”
Newson said hardware manufacturers should remove the possibility of a silent switch to software-based pseudo-randomness. She called for production systems to use one approved random-number generator, with the entire process from gathering entropy to creating a seed contained within a NIST FIPS 140-3-defined validation boundary.
“The strongest control is to remove the fallback from production and have exactly one approved RNG provider,” she emphasized.
She also said companies responding to an active attack must balance rapid patch development with clear communication. Users should be told immediately whether they are affected, what the vulnerability covers and what steps they can take, while any replacement firmware is tested thoroughly before release.
“The priority should be to immediately communicate the scope of the vulnerability, identify affected users, and provide clear mitigation guidance while thoroughly validating any fix before release,” Newson stated, adding that transparency is just as critical as the patch itself.
For users who hold compromised seed phrases and are concerned that an emergency firmware update could disable their devices, Newson recommended first obtaining a trusted hardware wallet and generating a new seed phrase offline. The new setup should then be checked with a small test transaction before all remaining funds are moved to it.
Only after those funds have been transferred should users consider updating the original device’s firmware, she said.
Newson also advised against concentrating every holding in hardware wallets made by one manufacturer. Splitting funds between devices from different companies, she said, can reduce the risk of a single point of failure.
The incident has intensified debate over the security assumptions behind self-custody. Critics argue that the sudden movement of dormant, long-term holdings demonstrates that keeping assets offline is not, by itself, an absolute guarantee of safety.
Nanak Nihal Khalsa, co-founder of Human.tech, said the exploit was another example of the third-party risks that remain within hardware ecosystems.
“‘Not your keys, not your coins’ misses an important fact: you are always outsourcing trust, even with self-custody. This just adds yet another point of evidence that self-custody does not change that fact,” Khalsa observed, warning that emerging threat vectors like AI-assisted exploits will likely compound these risks.
Newson said single-signature arrangements were especially exposed because one failure by a user, manufacturer or software dependency can result in the loss of an entire holding. She argued that broader adoption would require systems capable of limiting the damage caused by an individual mistake.
“Single-sig self-custody offers zero margin for error,” Newson said. “Users relying on a single device are trusting the physical hardware, the code, and all its dependencies as well as the QA checks to catch any issues.”
The industry is therefore increasingly considering multi-vendor, multi-signature and threshold signature, or MPC, arrangements as a minimum security standard.
“Yes, it should be the default baseline,” Newson concluded. “The goal is to move from ‘trusting a single device’ to ensuring that no single compromised component or actor can move the funds. In practice, signing keys or threshold shares should span independent organizational and technological failure domains, so that no provider can reconstruct the key or authorize a transaction alone.”
