Coldcard Exploit Update: Galaxy Links at Least 15 Attacker Clusters to Wallet Flaw

At least 15 separate attacker clusters have exploited a vulnerability affecting Coldcard hardware wallets, according to Alex Thorn, Galaxy Digital's head of research, after new victim reports surfaced and revealed additional, previously untracked thefts. Thorn said those disclosures enabled Galaxy researchers to tie more Bitcoin addresses to the incident. In one example, a victim who lost less than 1 BTC helped investigators identify an attacker that ultimately drained 12 BTC from 126 addresses. Galaxy noted that the pattern differs from a typical centralized exchange breach: funds were taken from many independently controlled wallets and dispersed to multiple attacker-controlled addresses. As a result, researchers may be unable to attribute or even detect some thefts without victims sharing wallet addresses and transaction histories. Galaxy Research estimates roughly $100 million in Bitcoin was stolen across three confirmed waves. A suspected fourth wave could push total losses to about $130 million. The firm emphasized that 15 attacker clusters do not necessarily equate to 15 identifiable individuals; some clusters may be controlled by the same actor, and additional attackers may remain undiscovered. Coldcard seed generation flaw reduced key randomness The issue centers on how some Coldcard devices generated recovery phrases. A firmware integration error reportedly caused affected units to rely on a deterministic pseudorandom number generator rather than the intended hardware-based random number generator. That reduced the effective randomness protecting certain private keys and made it more feasible for attackers to search for valid seed phrases. Castle Labs cofounder Francesco said impacted Coldcard wallets may have produced private keys with about 40 bits of entropy. By comparison, a standard 12-word recovery phrase is typically associated with 128 bits. The flaw did not allow attackers to extract keys directly from an air-gapped device. Instead, attackers could generate candidate recovery phrases and watch the corresponding Bitcoin addresses for incoming funds. Users who may be affected have been advised to update firmware, generate a completely new seed phrase, and move Bitcoin to addresses controlled by the new wallet. Simply importing an affected recovery phrase into another device does not fix the underlying weakness. AI testing debate reignited The incident has also fueled debate over whether AI-assisted security testing could have flagged the flaw earlier. Dragonfly managing partner Haseeb Qureshi argued that about "$2 of AI hardening" might have prevented the losses, citing experiments in which AI models reportedly rediscovered the vulnerability within minutes. One test claimed the open-source GLM 5.2 model found it in roughly 20 minutes without web access. Skeptics questioned whether those results demonstrate that AI could have identified the issue before it was public. Tokenomist data lead Tatsapat Saerejittima said some tests were conducted after disclosure and lacked blind testing, documented methodologies, and false-positive analysis. AI may lower the cost of finding vulnerabilities, researchers said, but it does not replace independent audits and end-to-end testing of how hardware wallets generate and safeguard private keys.