Cold Storage Is Not Cold: How a 2021 Firmware Bug Drained $130 Million From Coldcard Wallets

IT 7 Minutes

Cold storage was supposed to be the answer. Unlike exchange hot wallets — perennially connected to the internet, perpetually targeted by hackers — a hardware wallet lived offline, sealed inside a physical device, never touching a network unless its owner chose to sign a transaction. For years, the security pitch was almost unassailable: if your private keys are never online, they cannot be stolen online.

On July 30, 2026, that assumption broke in 41 minutes.

At roughly 03:00 UTC, a coordinated wave of transactions began sweeping Bitcoin out of thousands of Coldcard hardware wallets simultaneously. Within 41 minutes, 1,082 BTC — worth approximately $70 million at the time — had drained from 1,196 addresses and funneled into a consolidation wallet. Three more waves followed over the next four days. By August 4, blockchain analytics firm Elliptic had confirmed total losses exceeding $130 million across more than 5,200 compromised addresses. It was the largest hardware wallet exploit in cryptocurrency history, and it did not require a single device to be physically touched.

A Bug Born in 2021, Detonated in 2026

The flaw that made the Coldcard hack possible was not a clever new zero-day exploit. It was a configuration error introduced in firmware version 4.0.1, released in March 2021 by Coinkite, the Toronto-based company behind Coldcard. A single line of build configuration — setting MICROPY_HW_ENABLE_RNG to zero — had catastrophic consequences that would remain dormant for more than five years.

When that flag is set to zero, the device’s hardware random number generator (HRNG) is disabled during seed phrase generation. Instead of drawing entropy from dedicated silicon designed to produce true randomness, Coldcard’s MicroPython runtime silently fell back to a software pseudorandom number generator called Yasmarang, initializing itself with nothing more than the chip’s unique identifier, a timer register, and a handful of prior RNG call records. The result was seed phrases that appeared random but were, in practice, constrained to a narrow mathematical space that could be brute-forced without physical access to the device.

Security researchers later calculated the damage: Coldcard Mk2 and Mk3 devices running the affected firmware had an effective entropy of roughly 40 bits — compared to the 128 bits that proper BIP-39 seed generation requires. Mk4, Mk5, and Q-series devices fared somewhat better at approximately 72 bits, but still fell far short of the security threshold. An attacker who could narrow down a chip’s unique ID and approximate its timer state could generate candidate seed phrases offline, check each against public blockchain addresses, and drain matching wallets without ever touching the physical device or asking its owner for anything.

Coinkite released emergency patched firmware on July 31 — one day after the attack began. But the fix was, by definition, incomplete: updating firmware prevents new vulnerable seeds from being created. Seeds already generated on compromised firmware remain permanently weak. The company’s advisory was explicit: affected users must generate entirely new seed phrases on patched hardware and migrate all funds.

The Attack Itself: Coordinated, Multi-Wave, Multi-Actor

What happened on July 30 and the days that followed was not a single hacker running a script. On-chain analysis by TRM Labs identified distinct behavioral signatures across the four waves of theft, suggesting “multiple attackers may be involved” — likely independent groups who had each independently identified and exploited the same vulnerability, or who had purchased access to the exploit from a common source.

The first wave was the most dramatic: $70 million drained in 41 minutes, funds consolidated with surgical precision into a single staging wallet. The subsequent waves were smaller but methodical, eventually bringing the total confirmed stolen amount to approximately 1,816 BTC — roughly $116 million at July prices, and over $130 million as Bitcoin’s price continued rising into August.

Tom Robinson, co-founder and chief scientist of blockchain intelligence firm Elliptic, confirmed the $130 million figure and noted the involvement of “at least a dozen different hackers.” Despite the scale of the theft, on-chain laundering activity was surprisingly minimal: TRM identified a single 64.9 BTC deposit to Wasabi Coordinator (a Bitcoin mixing service) and 200 ETH deposited to Tornado Cash on August 4. Most stolen funds showed only one additional hop of consolidation before sitting dormant. Whether this reflects confidence, caution, or coordination among the attackers remains unclear.

The human cost became vivid when Jonathan Goodman, a software engineer who lost $1.6 million in Bitcoin, gave an interview to CBC News: “I did everything right. I never shared my seed phrase. The devices never touched the internet. I kept them in multiple safes and a safety deposit box. None of it mattered.” His story is emblematic of what made this incident particularly painful: the exploit required no operational security failures on the part of victims. Following best practices to the letter provided no protection against a flaw embedded in the device’s own firmware.

The Broader Hardware Wallet Security Crisis of 2026

The Coldcard exploit did not occur in a vacuum. July 2026 was already shaping up as a brutal month for crypto security before the Coldcard waves began. According to QuillAudits, total crypto losses in July reached $242 million across more than 30 incidents. The hardware wallet vulnerability alone — roughly 45% of July’s losses — outweighed the combined total of every DeFi protocol exploit that month, including a $24.15 million bridge attack against AFX, a $23.75 million oracle manipulation on Ostium, and a $21.2 million governance attack on BonkDAO.

Coldcard’s vulnerabilities landed alongside a separate but related threat to hardware wallet users: a wave of data breaches at shipping and fulfillment partners used by competing manufacturers. In August, both Trezor and SafePal disclosed that third-party shipping providers had suffered breaches that exposed customer names, home addresses, email addresses, and phone numbers. The devices themselves were not compromised in these incidents, but the stolen data created a different kind of risk.

Physical attacks on known cryptocurrency holders — sometimes called “wrench attacks” or “$5 wrench attacks” in security circles — have been escalating sharply. CertiK documented dozens of such incidents in 2025, representing a 75% increase over the prior year, with over $40 million stolen through physical coercion, kidnapping, and home invasions. Chainalysis estimates that physical crypto theft had already topped $30 million in the first half of 2026. Knowing that a home address belongs to someone who owns a hardware wallet creates targeting information that no firmware patch can address.

Self-Custody Redistributes Risk — It Does Not Eliminate It

The Coldcard exploit has forced a difficult conversation in the self-custody community. For years, the argument for hardware wallets rested on a simple premise: exchange hacks happen to other people; cold storage is different. The premise was never entirely accurate — hardware wallet vendors have faced targeted attacks before — but no single incident has so thoroughly dismantled the narrative that “if you hold your own keys, you are safe.”

TRM Labs put it plainly in their post-incident analysis: “The incident reinforces that self-custody redistributes rather than eliminates risk. Security depends on trustworthy key generation processes — firmware quality and entropy sources require ongoing scrutiny.” The same operational security discipline that protects against exchange hacks — knowing your counterparty, auditing their track record, understanding what you are trusting — applies with equal force to hardware wallet manufacturers.

The security community has long recommended multisignature setups as a defense in depth against any single point of failure. A 2-of-3 multisig arrangement using wallets from three different manufacturers, each with independently developed firmware, provides meaningful protection: even a complete compromise of one device’s key generation cannot drain funds without access to a second device’s signatures. The Coldcard incident gives this recommendation renewed urgency — and underscores that it should apply even to devices widely considered to be among the most secure on the market.

What This Means for You: A Practical Checklist

If you own or have ever owned a Coldcard hardware wallet, the following steps are critical:

  1. Identify your firmware version. The vulnerability affects Coldcard Mk2 and Mk3 devices running firmware versions 4.0.0 through 4.1.9, Mk4 and Mk5 devices running firmware before version 5.6.0, and Q-series devices running firmware before version 1.5.0Q. Check your device’s settings menu to see which version is installed.
  2. Do not simply update firmware. Updating firmware prevents new vulnerable seeds from being created, but it does not protect seeds already generated. If your wallet was set up on affected firmware, the seed is compromised regardless of what firmware version the device runs today.
  3. Generate a completely new seed on patched firmware. After updating to a non-vulnerable firmware version, perform a factory reset, generate a new seed phrase, and write it down on fresh paper or metal backup.
  4. Move all funds to the new wallet using test transactions. Before transferring your full balance, send a small test amount to confirm the new setup is working correctly. Verify receive addresses carefully.
  5. Check your old addresses on a block explorer. If your holdings on old addresses are still intact, move them immediately. If they have already been drained, preserve all transaction records and consult a legal advisor regarding potential recovery options.
  6. Consider multisignature custody going forward. For significant holdings, a 2-of-3 multisig setup using hardware wallets from different manufacturers substantially reduces the impact of any single device’s security failure.

If you are a Trezor or SafePal customer whose shipping data was exposed in the August breaches, be alert to unsolicited contact — phone calls, emails, or messages — from anyone claiming to be from the company, law enforcement, or a “recovery service.” Physical security awareness is also warranted: be cautious about who knows you hold cryptocurrency in your home.

A Wake-Up Call That Cannot Be Ignored

Jonathan Goodman followed every rule in the self-custody playbook and still lost $1.6 million. That is not an argument against self-custody — it is an argument for understanding what self-custody actually means. Holding your own keys transfers the security burden from an exchange’s engineering team to yourself, including the burden of evaluating whether the tools you use to generate and store those keys can actually be trusted.

The Coldcard incident began with a single configuration flag set to zero by a developer in 2021. It ended with over $130 million extracted from more than 5,000 wallets belonging to people who thought they had done everything right. The lesson is not that cold storage is a myth — it remains one of the most effective security postures available to individual cryptocurrency holders. The lesson is that cold storage is only as strong as the firmware, hardware, and key generation process that underpins it.

In a market where Bitcoin trades above $75,000 and where individual wallets can hold seven-figure balances, the economics of finding and exploiting entropy weaknesses have never been more compelling. The hardware wallet manufacturers that survive this era will be those who treat firmware security as a continuous, audited discipline rather than a solved problem. And the users who survive it will be those who hold their custodians — human and silicon alike — to the same standard.

Sources:

Unknown's avatar

Published by jamrockbrown

Published

Leave a comment