Share
A firmware flaw dating back to March 2021 has upended one of Bitcoin’s foundational promises. Beginning July 30, attackers began draining funds from Coldcard hardware wallets, eventually stealing roughly over 1,300 BTC, worth close to $89 million, from 4,585 addresses across multiple attack waves, according to Galaxy Research.
One victim lost $1.6 million in minutes despite following standard security steps. The exploit reignited a long-dormant debate, and Binance founder Changpeng Zhao (CZ) stepped into it with a claim that would have been heretical in crypto circles a decade ago: exchanges, he argued, are statistically safer than self-custody.
The vulnerability sat in the wallet’s random number generation. Certain Coldcard units, made by Canadian manufacturer Coinkite, fell back on a predictable software random number generator instead of the device’s dedicated hardware RNG when creating new wallets. That reduced the entropy, the randomness, feeding into each seed phrase, making it possible for attackers to reconstruct likely seed phrases offline and derive the private keys without ever touching the physical device.
.png)
No phishing, malware, or user error was involved. Instead, the device generated compromised cryptographic keys from the outset due to a software flaw that remained undetected for years.
The Coldcard hack is especially damaging to Canadian bitcoiners. Our analysis of attackers and victims finds that BTC holders in Canada are bearing 25% of attributable losses.
With estimates ranging as high as $110M, according to Galaxy Research’s dataset, we analyzed the… pic.twitter.com/AyxfHCcOrY
— Chainalysis (@chainalysis) August 4, 2026
Blockaid’s H1 2026 Onchain Security Report found that the first half of 2026 was the most active period for crypto exploits on record, with 212 verified incidents causing $1.1 billion in losses. While the number of attacks surged to 3.4 times the total recorded in all of 2025, losses remained below the previous year because no single breach matched the scale of the 2025 Bybit hack.
In H1 2026 alone, onchain incidents hit an all-time high, 3.4x of all of 2025.
Our team has worked alongside our customers and the industry to navigate these hacks.
We have also continued to strengthen our exploit detection platform, our product offerings, and deeply integrated… pic.twitter.com/VgUsHxrcgQ
— Idobn (@idobn) July 28, 2026
Compromised private keys accounted for nearly three quarters of all losses, while DPRK-linked actors were responsible for more than half of the stolen funds. The report also highlights emerging attack vectors targeting AI agents, EIP-7702 wallets, and bridge infrastructure, underscoring a shift toward increasingly sophisticated exploits.
Someone has attempted to contact the suspected COLDCARD attacker directly through the Bitcoin blockchain by sending a small “dust” transaction containing an OP_RETURN message. The message claims the sender can help launder approximately 562 BTC from the exploit, manage KYC requirements, and convert the funds into cash in exchange for a 10% commission. The sender spent only about $1.30 in transaction fees to deliver the proposal.
🚨COLDCARD THIEF JUST GOT A $1 ON-CHAIN OFFER TO LAUNDER 562 $BTC!
The main consolidation address sitting on ~562 $BTC from the Coldcard exploit received a dust transaction with an OP_RETURN message offering to clean the coins, handle KYC, and cash them out for a 10% cut.… pic.twitter.com/ZfrK3JLthS
— Crypto Banter (@crypto_banter) August 2, 2026
The message does not indicate that the stolen Bitcoin has moved or that the attacker has responded. The roughly 562 BTC remained untouched until Aug.2. At this stage, the transaction appears to be an unsolicited attempt to reach the holder of the stolen funds rather than proof of any laundering activity.
However, according to onchain tracking by TRM Labs, 64 BTC was sent to a crypto mixer, but only a small portion was processed in the first round. Most of the funds were returned as change before being divided into smaller chunks of roughly 7 BTC each for additional mixing attempts. Researchers note that these outputs remain unusually large, making the transactions relatively easy to follow on the blockchain and limiting the effectiveness of the laundering strategy.
Days after the exploit surfaced, CZ pointed to figures shared by analyst Willy Woo, sourced from River’s 2025 industry report, showing 1.57 million BTC lost through self-custody against 1.51 million BTC lost from exchanges, a gap under 60,000 BTC. “It is statistically safer to store coins on exchanges than to self-custody,” CZ wrote, while explicitly flagging that the conclusion depends on the underlying data being accurate.
It is statistically safer to store coins on exchanges than to self custody. 👇
A few thoughts:
Assuming the data is correct, my impression is Willy has always been a strong supporter for self custody.
Hack data is easier to collect on the CEX side, usually major news. It is… https://t.co/tqi1rUk5Q7
— CZ 🔶 BNB (@cz_binance) August 4, 2026
His reasoning for why the two figures might mislead in opposite directions is where the argument becomes more nuanced than the headline statistics. Exchange hacks are highly visible and typically well documented, making their losses easier to count. By contrast, self-custody losses, such as lost private keys, forgotten passwords, or destroyed hardware wallets, often go unreported, suggesting the available figures may understate their true scale.
Self-custody losses, lost private keys, forgotten passwords, destroyed hardware wallets rarely reach the same visibility, so CZ argued the self-custody figure likely understates the true scale of losses. He also noted that some exchange-side losses in Woo’s dataset come from platforms that no longer operate, meaning today’s surviving exchanges may carry a cleaner track record than the aggregate number suggests.
CZ also leaned on Binance’s own compensation infrastructure as evidence exchange custody has a real backstop that self-custody structurally cannot offer. The exchange has consistently covered user losses tied to CEX-side breaches through its Secure Asset Fund for Users (SAFU) recently expanded into a $1 billion Bitcoin reserve. A self-custody user who loses a seed phrase to a firmware bug has no equivalent fund to call.
CZ’s argument rests on aggregate Bitcoin losses rather than the causes behind those losses, but the history of crypto exchanges shows why many users remain skeptical of centralized custody. The industry’s largest failures were not isolated incidents. Mt. Gox collapsed in 2014 after losing around 850,000 BTC, while Bitfinex lost nearly 120,000 BTC in its 2016 breach. More recently, Bybit suffered a record $1.5 billion exploit in February 2025, underscoring that even modern exchanges remain attractive targets for sophisticated attackers.
The difference is that exchange losses are often socialized or partially recovered. Bitfinex eventually reimbursed affected users through BFX tokens, while Bybit replenished customer reserves within days following emergency financing.
Other exchange failures, however, ended very differently. Mt. Gox customers waited years through bankruptcy proceedings before receiving partial repayments, and FTX’s collapse highlighted that insolvency can be just as damaging as external hacks.
That distinction sits at the heart of the custody debate. Self-custody removes counterparty risk but leaves users fully responsible for protecting their private keys and hardware. Exchange custody shifts much of that operational burden to a third party, but requires trusting the platform’s security, governance, and financial health.
Rather than proving one model is inherently safer, the Coldcard exploit and the history of major exchange breaches suggest each custody method exposes users to a different set of failure modes.
The Coldcard exploit quickly shifted the discussion beyond the firmware bug itself, with security experts arguing that it should not be treated as evidence that self-custody is fundamentally less secure than exchange custody. Instead, they say the incident exposed a software supply chain and firmware integrity failure, not a weakness in the concept of holding your own private keys.
Jameson Lopp, co-founder and CTO of Casa, stressed that the breach highlights the importance of verifiable wallet software, open security audits, and layered protection rather than abandoning self-custody. Lopp has long advocated multisignature wallets and additional entropy sources to reduce dependence on a single hardware device or seed-generation process.
Willy Woo, whose Bitcoin loss data CZ referenced, also rejected the idea that the exploit invalidates self-custody. Responding to questions about Coldcard’s “available source” software model, he noted that users ultimately have to trust someone, whether it is a wallet manufacturer or a custodian. His preferred approach remains multisignature custody, which distributes signing authority across multiple keys instead of relying on a single seed phrase.
Security specialists have also pointed to a key technical lesson from the incident: updating Coldcard’s firmware alone does not protect affected wallets. If a seed phrase was created using the vulnerable firmware, the only effective mitigation is to generate a completely new seed with the patched software or another trusted environment and migrate funds. Simply importing the old seed into another hardware wallet does not eliminate the underlying weakness because the compromised entropy remains embedded in the original seed phrase.
Regardless of who wins the statistical argument, onchain behavior gives a clearer signal than either side’s perspective.
CryptoQuant data shows daily Bitcoin deposits to exchanges in transactions under 10 BTC jumped to 7,300 BTC on July 31, the highest level since February 6, while daily active addresses spiked from 645,000 to nearly one million, the highest since December. That is a striking reversal from the pattern after FTX’s 2022 collapse, when investors rushed to withdraw coins from centralized platforms into self-custody. This time, fear is flowing in the opposite direction, toward exchanges rather than away from them.
Coldcard users are fleeing at FTX-collapse speed.
Sub-1 BTC transfers hit 39.6K BTC on July 31, days after the hack disclosure.
That's just shy of the 39.9K BTC moved when FTX collapsed in Nov 2022. Small holders haven't moved at this scale in nearly 4 years. pic.twitter.com/g5AIgsOpLM
— CryptoQuant.com (@cryptoquant_com) August 4, 2026
CZ himself stopped short of recommending exchanges over private wallets outright, instead framing custody as a matter of risk tolerance and product fit rather than a simple safety ranking, and separately floated wallet diversification, not concentrating funds behind a single seed-generation process, firmware version, or wallet model, as a practical hedge regardless of which side of the debate one lands on.
Industry-wide hack incidents climbed roughly 50% in the first half of 2026 even as total stolen sums fell, a pattern that suggests attackers are shifting toward smaller, more numerous compromises rather than singular mega-breaches.
Both custody models are absorbing that shift. Neither has produced a clean statistical winner, and the Coldcard exploit’s real lesson may be that the debate itself is built on a false binary: the question was never simply “exchange or self-custody,” but which specific, auditable failure mode each choice exposes a holder to.
Share
preferred on
