Coldcard Seed-Generation Flaw: BTC Drained From Offline Wallets, AI Red Team Finds Thousands of Issues
- Filed
- Cryptocurrency
- First run
- 2026-08-22
- Length
- 1658 words
Correction and update (November 1, 2026): Coinkite's official advisory (published July 30, 2026, continuously updated) confirms: seeds generated on affected firmware from March 2021 through July 2026 are at risk — Mk2/Mk3 seeds from firmware 4.0.1–4.1.9 are at risk unless supplemented with at least 50 fair dice rolls and a strong BIP-39 passphrase; Mk4/Q/Mk5 seeds generated before the fixed firmware carry roughly 72 bits of entropy versus the expected 128. The confirmed loss accounting (Galaxy Research): three attack waves totaling ~1,367 BTC (~$88.6M) by August 2, with later media estimates climbing past ~$110M; this article's original "1,816 BTC," "~40-bit entropy," "four lines of code," and "15 attacker groups" figures were never confirmed by official or traceable sources and have been corrected in the body. The red-team numbers (390 projects / 27.5 hours / 4,962 findings) are corroborated by multiple outlets. Firmware updates do not repair an already-generated weak seed — migrate to a new seed per the official guidance.
Verification note — September 15, 2026
Prices, plans, eligibility, regional availability, fees, USSD codes, network costs and service status in this article can change without notice. This review uses Coinkite security advisory as the official recheck point. Historical figures and hands-on observations remain dated context, not a current quote, availability promise, or payment-success guarantee. Before buying, topping up, moving assets, or sending an on-chain transaction, recheck region, tax, network and expiry on the provider’s final checkout, app/support channel, or wallet confirmation screen.
What happened: weak seeds, eight figures in bitcoin gone
Starting July 30, 2026, users of the Coldcard hardware wallet (made by Coinkite) watched bitcoin drain from addresses whose keys never touched an internet-connected device. Multiple independent attacker groups swept the chain in parallel — Galaxy Research identified at least three waves, with a possible fourth. Its on-chain accounting: the first wave pulled ~1,082 BTC from 1,196 addresses in 41 minutes; by August 2, three waves totaled ~1,367 BTC (about $88.6 million) across more than 4,300 addresses, and later media estimates climbed past the $110M mark — the largest hardware wallet exploit of the year.
The root cause was an RNG integration error introduced in a March 2021 firmware update: entropy generation that should have used the hardware true random number generator fell back to a deterministic software PRNG (Block's engineering team traced it to ngu.random falling back to MicroPython's Yasmarang instead of the STM32 hardware RNG). Per Coinkite's advisory: Mk2/Mk3 seeds from firmware 4.0.1–4.1.9 are at risk unless supplemented with at least 50 fair dice rolls and protected by a strong BIP-39 passphrase, and Mk4/Q/Mk5 seeds generated before the fixed firmware carry about 72 bits of entropy versus the expected 128. Weak seeds can be enumerated offline and matched against public addresses on the ledger — no device access required. Your cold wallet could sit in a drawer while the coins left.
What compressed entropy means in practice
Start with the official numbers. BIP-39 seeds are designed for 128-256 bits of entropy; the affected Mk4/Q/Mk5 seeds carried about 72 bits, and Mk2/Mk3 seeds without dice-roll supplementation and a passphrase were worse. Seventy-two bits still sounds like a lot, but it gives up a factor of 2^56 in safety margin against the design target — and the attack runs offline at scale: generate candidate seeds, derive addresses, compare against public addresses on the ledger, around the clock, without touching a single device. The multiple sweep waves that followed the July 30 disclosure prove the route works.
And it sat undetected for five years, because low entropy produces no functional symptom. Wallets generated fine, signed fine, spent fine. The only way to catch it is a reverse audit of the generation path — precisely the kind of grinding code review AI now excels at and humans avoid.
The AI red team: 390 repos in 27.5 hours
After the breach, AnchorWatch CEO Rob Hamilton and Bitcoin developer Calle assembled a 16-person "Bitcoin Red Team" with a simple premise: attackers are using AI to find bugs; defenders must use AI to find them first. The sprint's output:
- 27.5 hours, 390 open-source Bitcoin projects scanned
- 4,962 findings filed — 85 critical, 635 high-severity
- About three critical findings per hour, team-wide (85 in 27.5 hours)
- Cryptography libraries were the epicenter: 1,385 issues across 128 projects, over a quarter of the total sample
By category, mining pools showed a 21.7% finding rate, infrastructure/tooling 21.5%, swaps and exchanges 20.9%, privacy tools 24% (from only 3 reviewed projects). Curiously, hardware wallet firmware — Coldcard's own category — sat second-lowest at 9.6% critical share, yet produced the only real-world exploit. Calle's summary: "decades of open-source code colliding with a two-week-old Kimi K3," and "everything is broken, Bitcoin is on fire." The sweep kept growing: by August 13, 25 developers had covered 501 projects over roughly 108 hours, filing 7,958 findings (1,280 rated high or critical), with OpenSats funding the compute.
The actual story: defenders can't get the tools
The red team originally planned to run on OpenAI and Anthropic models. It didn't work out: OpenAI restricts cybersecurity-related usage, and Hamilton — after completing full identity verification — had his account banned within 20 minutes of using the model to analyze a public codebase. The team pivoted to Moonshot AI's Kimi K3 (with OpenAI systems and Anthropic's Claude variants also in the mix per media accounts, always with human verification of AI leads; OpenSats funded compute costs).
The asymmetry is stark: attackers face zero restrictions on any tool they can rent; verified defenders get rate-limited, throttled, or banned. On August 10, the Bitcoin Policy Institute coordinated an open letter — "Defenders Need the Frontier" — signed by 40+ organizations including Coinbase, Block, BitGo, Blockstream, Trezor, Anchorage Digital, ARK Invest, Chaincode, Brink and Foundry (the signatory list later grew to roughly 80). It doesn't ask for unrestricted access: it proposes vetted access programs with pre-release models, compute, private code-review sandboxes, and direct channels to lab security teams.
There were reports that decentralized exchange Boltz paused bitcoin swaps (not confirmed by any official announcement) — a sign of AI-driven discovery outrunning patching capacity. When the find-rate exceeds the fix-rate, that gap is itself a systemic risk.
Timeline of the incident
- Mar 2021: Coldcard firmware update ships the RNG integration error (later traced by Block's engineering team).
- Jul 30, 2026: Anomalous withdrawals appear; Coinkite publishes an emergency advisory; multiple attack waves begin (Galaxy counted ~1,196 addresses hit in the first 41 minutes).
- Early Aug 2026: The 16-person red team forms; 390 projects and 4,962 findings in the first 27.5 hours. Hamilton's verified OpenAI access is banned in 20 minutes; the team shifts to Kimi K3.
- Aug 10, 2026: "Defenders Need the Frontier" open letter published with 40+ signatories.
- Mid-Aug 2026: Loss estimates keep climbing (by agency and cutoff; Galaxy's Aug 2 accounting stood at ~1,367 BTC / $88.6M); the red team expands to 501 projects and 7,958 findings.
Two caveats worth keeping honest: loss estimates rose over time and differ by methodology and cutoff — Galaxy Research's August 2 accounting stood at three waves totaling ~1,367 BTC (~$88.6M), with later media totals past ~$110M — and the 4,962 "findings" are AI leads with human triage, not 4,962 confirmed exploitable vulnerabilities (only ~24.7% of the expanded 7,958 findings had reproducible proofs). The numbers will be revised; the structural problem won't.
What this means for anyone holding digital assets
- "Open source = secure" is dead. Coldcard was fully open — "anyone can read the code" — and a five-year entropy bug survived until attackers monetized it. Open source's value now depends on whether AI can read it faster than attackers do.
- Key generation is the softest spot. The bug wasn't in transmission or signing; it was in the instant the secret was born. The same applies to your card CVVs, API keys, and wallet seed phrases: whatever is determined "at generation time" deserves independent verification, and generation should never co-locate with storage.
- Don't single-source custody. Losses concentrated among users who didn't move funds promptly. Split meaningful balances across independent key origins rather than betting on any vendor's firmware hygiene.
A three-minute self-audit
- Coldcard owners: check firmware version and when your seed was generated. Seeds created after the March 2021 firmware on affected batches must be treated as burned — move funds and re-generate with fresh entropy (Coinkite's security notice has the official guidance).
- All hardware wallet users: have you done a small-amount test recovery from your seed? A successful restore validates both your backup and your possession of the key.
- Virtual card users: are your top-up address, bound email, and 2FA channel independent? One compromise should not cascade into all three.
- Subscribe to your platforms' security disclosure channels — in this incident, users who saw the Coinkite advisory on day one kept their funds; losses concentrated among the late-informed.
FAQ
Is Coldcard still usable? New firmware is patched; seeds generated in the affected window are permanently compromised and must be regenerated.
Can the stolen BTC be recovered? Realistically no — multiple waves and ongoing mixing; this is the canonical "hacked while offline" case.
Is AI auditing just false-positive spam? The pipeline is AI-leads-plus-human-triage; 85 critical findings in 27.5 hours is a density signal, not a confirmed-exploit census.
Does this affect exchanges? The red team's scan showed a ~20.9% finding rate for swaps/exchanges; funds on a CEX add another layer of patch-speed dependency.
Bottom line
Three lessons from the Coldcard incident. Technically: entropy generation is the root of key safety, and a single RNG integration error ended the myth that cold storage equals absolute security. Industrially: AI has made vulnerability discovery exponentially faster while defenders are throttled by their own compliance — "Defenders Need the Frontier" is a survival statement, not a slogan. Personally: whether you manage an eight-figure wallet or a handful of virtual card balances, independent key generation, distributed custody, and fast response to security advisories remain the three habits always worth your time.