Coldcard firmware bug triggers shift to multisig security
- A Coldcard randomness bug allowed attackers to steal 2,000 Bitcoin across 1,200 wallets.
- Black-hat attackers used AI tools to quickly find the flawed firmware code.
- Developers are abandoning single-device setups in favor of multi-vendor multi-signature vaults.
Single-device Bitcoin self-custody broke down overnight. A severe entropy flaw in Coldcard Mark III hardware wallets allowed automated attackers to drain roughly 2,000 Bitcoin across 1,200 wallets. Users who believed they followed industry best practices watched their life savings disappear into visible address clusters. The disaster upended years of dogmatic trust in single hardware vendors.
The mechanics were shockingly simple. On What Bitcoin Did, host Danny Knowles hosted Frostsnap developers Lloyd Fournier and Nick Farrow to unpack how manufacturer CoinKite failed. CoinKite bypassed standard cryptographic protocols, routing randomness through a toy algorithm named Yasmerang written in MicroPython. Device key security collapsed from the standard 256 bits down to roughly 20 bits. That reduced the key search space from trillions of galaxies to a few hundred million possibilities.
Automated AI models turned a theoretical vulnerability into a rapid execution. On TFTC, host Marty Bent discussed the breach with Arc Labs guest Alex Bergeron, who noted that black-hat agents used large language models to map the flawed code across open-source repositories. Automated scripts quickly scanned the blockchain and swept target addresses. Small security teams could not patch the flaw before automated tools exploited it.
The attacker left a trace during the heist. On Bitcoin And, host David Bennett detailed how Block engineering lead Clay Garrett linked the thief to subscription logs at a commercial data vendor. The attacker queried target addresses using a paid account right before executing the sweeps. Block forwarded matching subscriber records to federal law enforcement, though recovering irreversible protocol-level transactions still requires securing private keys.
The fallout ignited a fierce debate over hardware design philosophy. On BTC Sessions, wallet developers from Blockstream, Trezor, Foundation Devices, and SeedSigner clashed over how to fix key generation. Developers noted that users who manually generated entropy using physical dice rolls avoided the bug entirely. SeedSigner argued that hardware devices should mandate manual entropy generation. Trezor and Blockstream counter-argued that extreme manual friction drives average users into centralized custodial ETFs.
Hardware vendors are now adjusting to a world where AI speeds up vulnerability discovery. Developers across Foundation Devices and Trezor acknowledged that open-source code visibility is no longer sufficient without continuous active auditing. Wallet teams are integrating automated AI code reviews into firmware release pipelines to catch flaws before external bad actors scan their repositories.
Relying on a single hardware manufacturer is no longer considered acceptable self-custody practice. Fournier and Farrow advocated for collaborative randomness protocols like Frostsnap, which split entropy generation between a phone and a hardware chip. Meanwhile, users who held funds in multi-vendor multi-signature arrangements survived the Coldcard flaw without losing satoshis or needing emergency key migrations.
The ecosystem is shifting toward smart co-signing structures that remove single points of failure. Bergeron pointed to Arc Labs' Arcade Vaults, which combine mobile passkeys, automated server co-signers, and vault operators like River to protect funds while maintaining recovery paths. Rather than isolating users in fragile DIY setups, modern multi-signature architectures enforce spending caps and safety nets without relinquishing final sovereignty.
Self-custody is no longer defined by trusting a single piece of hardware in a safe. True sovereignty requires distributed keys, independent software stacks, and architectural redundancy.