After $70M Coldcard Exploit, CZ Urges Split Custody for Bitcoin: ‘Nothing Is 100%’

When a hardware wallet’s safety net fails, it fails fast. In a 41-minute window on July 30, roughly 1,082.65 BTC—about $70.2 million—was swept from 1,196 Bitcoin addresses linked to a Coldcard seed-generation flaw. Binance founder Changpeng “CZ” Zhao responded with a reminder many ignore in bull cycles: nothing in custody design is absolute. His ask was simple—avoid single points of failure by splitting funds across multiple wallets—while acknowledging the operational trade-offs that come with that choice.

The critical issue: entropy assumptions. A build error in firmware shipped in March 2021 caused some Coldcard devices to draw seed entropy from a software fallback rather than the hardware random-number generator. That subtle shift collapses the search space for keys, turning “cold” into guessable. Importantly, updating firmware cannot harden a seed that was already created under the faulty path; those seeds have to be replaced and funds moved.

Galaxy Research, building on a transaction pattern identified by engineers at Block and shared by @clay_garrett, mapped the damage: 1,196 addresses drained in full between 01:10:20 and 01:51:26 UTC. Early tallies had pegged losses near 594 BTC (~$38 million) across around 500 wallets; the expanded analysis nearly doubles that. The on-chain signature suggests automation: each sweep used an identical hardcoded fee, produced no change output, and quickly consolidated into a handful of addresses that, as of now, have not moved. Victims spanned native SegWit and older address types, pointing to multi-path key scanning rather than ad hoc manual theft.

Coinkite has shipped emergency hotfixes and urged exposed users to migrate to newly generated seeds. That is necessary, but users often underestimate the practical complexity of response: inventorying affected backups, coordinating secure moves, and avoiding self-inflicted mistakes under time pressure. This is where design philosophy matters more than brand: assume components fail, and architect so failure in one layer does not cost the treasury.

How to internalize this:

- Diversify key origin. Split capital across independent wallets and ideally across vendors and firmware lines. A single RNG or build system should not be able to jeopardize everything.

- Consider multi-vendor multisig (e.g., 2-of-3) for meaningful balances. One compromised signer should not allow spend. Keep quorum devices isolated and updated, and rehearse recovery.

- Strengthen entropy. Where supported, add user-supplied randomness (e.g., dice) to hardware-generated entropy. Verify seed creation steps offline and document provenance.

- Treat “known-bad window” seeds as tainted. If a seed may have been created on affected firmware, move funds to a fresh seed generated on a remediated device; a firmware update alone won’t fix the old seed.

- Monitor clusters and automate alerts. Set spend alerts on custody addresses and predefine an emergency playbook to accelerate moves without improvisation.

- Accept the trade-offs. Splitting funds increases operational overhead and human-error risk. Balance complexity with clear SOPs, regular drills, and minimal attack surface.

There’s also a responsibility shift for manufacturers. Deterministic, reproducible builds, continuous entropy self-tests, third-party audits, and rapid, transparent incident communication reduce blast radius and rebuild confidence. Users, meanwhile, can resist a common cognitive trap: assuming brand longevity equals invulnerability. Long histories help, but as this event shows, a single build regression can ripple for years.

CZ’s message lands because it is pragmatic rather than maximalist. Cold storage remains valuable, but cold is a posture, not a product. If a single bug can take you to zero, the design—not just the device—needs an upgrade. Stay informed, rotate compromised seeds, and keep your setup SAFU—by construction, not by hope.