What the Coldcard Hack Means for Self-Custody

John.H·Aug 5, 2026·8 min read

The Coldcard firmware exploit drained over $100M in Bitcoin, exposing the real risks behind hardware wallet security and self-custody.

On July 30, 2026, an attacker began draining Bitcoin from Coldcard hardware wallets without ever touching a single device. Within 41 minutes, approximately 594 BTC had been swept from roughly 500 dormant addresses. Over the next five days, three more waves followed. By August 4, Galaxy Research confirmed that 1,596 BTC had been stolen from more than 7,300 addresses, with confirmed losses exceeding $100 million. Fortune and Forbes reported total losses including suspected theft at approximately $116 million, and the number may still be growing.

The incident is the largest hardware wallet exploit in Bitcoin's history. It has forced the crypto industry to confront an uncomfortable question: if cold storage can fail this catastrophically, what does secure self-custody actually look like?

How the Firmware Flaw Worked

The vulnerability traces back to a single code change made on March 1, 2021. A firmware update for Coldcard's Mk3 device altered how the wallet generated seed phrases, the 12 or 24 word sequences that serve as the master key to a Bitcoin wallet. The change was subtle. The seed generation function was redirected from the device's dedicated hardware random number generator to a software-based substitute built into the MicroPython runtime.

The problem was that the software fallback was never meant to be used for cryptographic key generation. It initialized its randomness from the chip's unique serial number and internal clock readings, neither of which is secret. After initialization, it collected no additional entropy. The result was that seeds generated on affected devices contained roughly 40 bits of effective randomness instead of the expected 128 bits.

To put that in perspective, 128 bits of entropy would require approximately 340 undecillion guesses to brute force. Forty bits requires roughly one trillion guesses, a computation that modern hardware can complete in hours. The attacker did not need physical access to any device. They reconstructed candidate seeds offline by modeling the predictable inputs, derived the corresponding Bitcoin addresses, matched them against the blockchain, and swept the funds.

The vulnerability affected Coldcard Mk3 devices running firmware versions 4.0.1 through 5.0.3, a window spanning from March 2021 through the final Mk3 firmware release. Mk4 and newer models were also affected, though to a lesser degree, with estimated entropy of approximately 72 bits instead of 128, enough to make brute-force attacks computationally expensive but not theoretically impossible.

Coinkite, the company behind Coldcard, pushed emergency firmware on July 31 and halted all shipments. The company subsequently destroyed its remaining inventory of devices manufactured with vulnerable firmware and began coordinating with law enforcement. But the patched firmware only protects newly generated seeds. Any wallet created during the five-year vulnerability window remains compromised until the user generates a fresh seed on updated firmware and transfers their funds.

The Scale of the Damage

Galaxy Research, which has been tracking the exploit in real time, identified four distinct attack waves between July 30 and August 3. The first wave was the most concentrated, draining 1,082.65 BTC from 1,195 addresses. Subsequent waves targeted progressively more addresses as the attacker expanded their search across the pool of vulnerable seeds.

As of August 4, Galaxy is working with 73 individual victims and reports that approximately 90% of the stolen coins remain static and traceable on chain, offering some hope for recovery. But the investigation is still ongoing, and no arrests have been made. While recent large-scale crypto thefts have often been attributed to state-backed groups, investigators have not yet linked this incident to any specific actor.

The attack has raised concerns that additional vulnerable wallets may still hold funds. Coinkite's advisory warned that every single-signature wallet created on an affected device without a user-supplied BIP-39 passphrase or manual dice roll entropy is potentially at risk. Users who added a passphrase or rolled at least 50 dice during setup introduced external randomness that the attacker cannot reconstruct, making those wallets safe.

What This Reveals About Self-Custody Risk

The Coldcard exploit did not involve phishing, social engineering, or supply chain tampering. It was a manufacturing defect in the most security-critical function a hardware wallet performs: generating the private key. The failure occurred at the moment of wallet creation, and the user had no way to independently verify that the randomness was insufficient. Every affected owner followed the correct procedure. They purchased a reputable device, generated a seed offline, wrote it down, and stored it securely. The vulnerability was invisible to them.

This distinction matters because the crypto industry has historically framed self-custody as a binary choice. Either you control your keys and are safe, or you trust a third party and are exposed. The Coldcard incident demonstrates that self-custody carries its own category of risk that is fundamentally different from, but not necessarily smaller than, exchange custody risk.

Security researcher Lorenzo Valente described the tradeoff clearly: self-custody replaces counterparty risk with software risk, hardware risk, supply chain risk, and phishing risk. Each of these failure modes is real and has resulted in documented losses. The Coldcard case falls squarely into the hardware and software risk category.

Survey data from 2026 illustrates the gap between perception and practice. Approximately 57% of crypto holders say they prefer self-custody, and 46% report fearing a major exchange breach. Yet 88% of those surveyed still keep the majority of their assets on centralized exchanges, and only 33% use a cold wallet. The preference for self-custody is aspirational. The actual behavior reflects an intuitive understanding that managing your own security is difficult.

Exchange Custody Is Not the Answer Either

None of this means that exchange custody is inherently safer. The history of centralized exchange failures is extensive and well documented.

The Bybit hack in February 2025 resulted in the theft of approximately 401,000 ETH, worth $1.5 billion, in a single incident. The FBI attributed the attack to North Korea's Lazarus Group. It remains the largest single theft in cryptocurrency history. Mt. Gox lost approximately 850,000 BTC between 2011 and 2014 due to sustained exploitation of weak internal controls. FTX's collapse in November 2022 exposed an $8 billion shortfall in customer funds, with an additional $477 million drained from wallets during the bankruptcy process.

Annual losses from crypto hacks and exploits are estimated between $3 billion and $14 billion depending on the methodology and scope of the analysis. The first half of 2026 alone saw more than $1 billion in losses across the industry.

Exchange custody concentrates risk in a third party. If the exchange is hacked, goes bankrupt, or freezes withdrawals, users are dependent on legal proceedings that can take years. The Mt. Gox creditor repayment process took more than a decade. Self-custody concentrates risk in the user's own hardware, software, and operational discipline. The two risk profiles are different in kind, not in magnitude.

Hardware Wallet Security After Coldcard

The Coldcard incident has also highlighted vulnerabilities in other hardware wallet ecosystems, though none have resulted in comparable losses.

In January 2026, attackers compromised the payment processor used by Ledger, stealing customer names, addresses, email addresses, and phone numbers. Financial data and seed phrases were not affected, but the breach fueled phishing campaigns in which victims received physical letters with QR codes prompting them to enter their recovery phrases.

In June 2026, Ledger's Donjon security team identified a laboratory attack against the TROPIC01 chip used in Trezor's Safe 7 wallet. Trezor disclosed the vulnerability and stated that it does not put user funds at risk in practice, as exploitation would require physical access, specialized equipment, and advanced expertise.

These incidents illustrate that hardware wallet security is not a single property but a stack of interdependent layers: chip integrity, firmware correctness, random number generation, supply chain authentication, and the manufacturer's data security practices. A failure at any layer can compromise the entire system, even if every other layer functions perfectly.

What Users Should Do

For Coldcard owners, the immediate priority is clear. Anyone who generated a wallet on a Mk3 device running firmware 4.0.1 or later, without adding a BIP-39 passphrase or performing a manual dice roll, should assume their seed is compromised. The recommended steps are to update to the latest firmware, generate a completely new seed on the updated device, and transfer all funds from the old wallet to the new one.

For the broader market, the Coldcard exploit reinforces several principles that apply regardless of which hardware wallet you use.

Multisignature setups, which require multiple independent devices to authorize a transaction, would have prevented this exploit entirely. Even if one device generated a weak seed, the attacker would need to independently compromise each signing device. The added complexity of multisig is the cost of eliminating single points of failure.

Passphrases add entropy that is independent of the device's random number generator. A strong passphrase transforms a potentially weak seed into one that cannot be reconstructed without knowledge that exists only in the user's memory or secure backup.

Verification matters. Open-source firmware allows the security community to audit the code that generates and protects private keys. The Coldcard flaw went undetected for five years in part because the specific code path involved was not thoroughly reviewed after the March 2021 change.

The Coldcard hack has not invalidated self-custody as a concept. But it has demonstrated that self-custody is not a passive state. It requires ongoing attention to firmware updates, hardware advisories, and evolving best practices. The promise of self-custody is sovereignty over your own assets. The price of that sovereignty is the responsibility to actively maintain the security infrastructure that protects them.