In what is quickly becoming one of the most devastating breaches in cryptocurrency history, a catastrophic Coldcard wallet hack has resulted in the theft of over $116 million in Bitcoin. The exploit, which targets premium devices manufactured by Canadian security firm Coinkite, shatters the bedrock assumption that offline hardware wallets are impenetrable. More alarmingly, the company issued a stark warning to the broader tech sector in the aftermath: advanced artificial intelligence failed completely to flag the fatal code before deployment, leaving thousands of investors exposed.
The $116 Million Bitcoin Drain: How It Unfolded
The sheer scale of the theft became apparent in late July 2026. Blockchain intelligence firm Galaxy Research reported that an attacker moved approximately 1,816 Bitcoin—worth roughly $116 million—out of more than 5,200 addresses generated on Coldcard devices. The funds were systematically drained in at least four organized waves, with the transaction sweep rate hitting an unprecedented 45 times the pre-incident baseline. Early estimates suggested $30 million in losses, but figures rapidly ballooned as hackers broadened their scope to more addresses following the initial Coinkite Bitcoin exploit.
Behind the raw blockchain data are devastating personal losses. Canadian entrepreneur Jonathan Goodman reported losing 18.25 Bitcoin, equivalent to CAD $1.6 million, despite keeping his Coldcard device strictly secured in a physical safety deposit box. His wallet never once connected to the internet, proving that a severe crypto cold storage vulnerability had fundamentally compromised the devices from the moment they were initialized. The attacker's systematic approach suggests multiple groups are now racing in parallel to drain remaining vulnerable addresses.
Dissecting the Hardware Wallet Seed Phrase Bug
Security researchers at Block traced the root cause of the exploit to a catastrophic integration error deeply embedded in the device's firmware. Specifically, a hardware wallet seed phrase bug was introduced in firmware version 4.0.0, released over five years ago in March 2021.
When affected users initialized their wallets, a faulty configuration bypass flag instructed the device to ignore its dedicated STM32 hardware random number generator. Instead, the firmware routed seed generation to a deterministic software substitute known as MicroPython's Yasmarang pseudorandom number generator. Because this fallback mechanism collected no fresh cryptographic entropy and relied entirely on predictable, static variables like the device identifier and system timer state, the resulting seed phrases were severely constrained.
Why Keeping Devices Offline Failed
Hackers did not need to physically access the devices, employ phishing tactics, or trick users into revealing their credentials. They utilized the predictable mathematical constraints to reproduce candidate output streams offline, deriving addresses and checking them against the public blockchain. This devastating Coinkite Bitcoin exploit effectively reduced the cryptographic security of a 12-word seed phrase from a robust 128 bits down to highly vulnerable levels, making the keys guessable for anyone with sufficient computing power.
A Catastrophic AI Code Review Security Failure
Beyond the immediate financial damage, the Coldcard wallet hack exposes a massive, systemic blind spot in modern cybersecurity workflows. Coinkite CEO Rodolfo Novak revealed that the company ran top-tier AI models over their critical codebases just weeks prior to the exploit. The AI code review security failure was absolute: the automated models detected neither this specific random-number bug nor any other severe vulnerabilities in the firmware architecture.
Coinkite explicitly warned other cryptocurrency projects that while defensive AI tools failed to protect their platform, malicious actors are likely leveraging the exact same technology to aggressively scan open-source repositories for similar flaws. The incident highlights a dangerous asymmetric advantage for attackers. AI can rapidly surface obscure dependencies, subtle logic flaws, and fallback errors in open-source histories that both human reviewers and standard defensive AI audits routinely miss. The company cautioned developers to assume that public firmware is constantly being read by attackers, urging immediate manual audits of build boundaries.
Bitcoin Self Custody Risk and Immediate Next Steps
This exploit fundamentally redefines the concept of Bitcoin self custody risk. For years, the industry standard dictated that keeping a hardware device completely air-gapped was the ultimate defense against cyber theft. However, as this event proves, physical isolation is entirely meaningless if the underlying cryptographic entropy is inherently flawed. The trustless model of digital assets faces extreme scrutiny when a single point of technical failure can wipe out years of savings.
Coinkite released an emergency firmware update for all models on July 31, but installing the patch does not repair seeds that were already generated using the compromised code. Users currently holding assets on devices configured with the vulnerable firmware must take immediate, decisive action. The only secure mitigation strategy is to update the device to the newly patched firmware, generate an entirely new seed phrase, and broadcast transactions to transfer all remaining funds to the newly secured addresses. For absolute certainty and peace of mind, security experts and Coinkite highly recommend bypassing automated generation altogether by using the device's manual dice-roll feature—rolling at least 50 independent, physical dice to generate completely verifiable, trustless entropy.
