Was AI Found the Coldcard Bug? $70M in BTC Says the Search Just Got Faster

The Coldcard loss magnitude changed the story

This stopped being a theoretical security warning when over 1,082 BTC, around $70 million was pulled from 1,196 Coldcard wallets in a single 40-minute window. That scale makes it more than a product-specific incident: it is a liquidity event with implications for confidence across self-custody.

AI is an expert read, not a confession

The AI angle is being discussed cautiously. Reports say industry experts believe AI was used in the breach, while Block's engineering team said the transaction trail identified the thief but did not identify the vulnerability behind it. The cleaner takeaway is to treat AI as a likely part of the attacker's toolkit, not as a settled explanation of how the bug was discovered.

Why Coldcard mattered beyond a routine exploit

Coldcard was not a fringe tool. Coinkite is one of the most respected hardware wallet manufacturers, and the flaw sat in firmware 4.0.0 through 4.2.0, affecting seeds created since March 2021. When a trusted self-custody brand has a randomness failure, the shock spreads quickly because it challenges the assumption that hardware wallets automatically mean safe.

What likely happened: a weak fallback, not a broken Bitcoin protocol

The core issue was not a break in Bitcoin cryptography. A build configuration error in firmware version 4.0.0 replaced Coldcard's secure hardware random number generator with a predictable software-based fallback that used non-secret chip data. In practical terms, affected wallets still looked normal, but the randomness behind seed generation was much weaker than intended.

Entropy, not Bitcoin, was the failure point

The problem was not that seeds became trivially guessable. It is that the keyspace shrank enough to become tractable for automated brute-force search. On affected Mk3 devices, effective entropy fell to roughly 40 bits. Mk4 and Q devices were less exposed, at roughly 72 bits, but still below the 128 or 256 bits the crypto-security community considers standard. That helps explain why automated enumeration could be so effective.

Not every Coldcard user was equally exposed

This was not a sweep of every Coldcard user. The issue traced to seeds created during a window starting in March 2021, tied to firmware 4.0.0–4.2.0. It hit specific generation paths hardest: 12- or 24-word seeds without dice rolls or a BIP 39 extra passphrase were the vulnerable case. That keeps the event from looking like "Bitcoin is broken" and points more directly to a subset of wallets with a smaller keyspace.

Why AI sounds plausible, and where the evidence stops

The scale and speed of the drain make automation look central. Reports track 1,082.65 BTC moved from 1,196 addresses, with most funds shifting in a 40-minute window. That fits automated enumeration of a weakened keyspace. Commentary has also suggested AI was involved. The cautious read is that AI or similar automation may have helped the search move faster, not that the reports yet prove exactly who attacked or how the bug was initially found.

What to do next, and why the wider ecosystem should pay attention

The timing is still the clearest warning. The majority of funds moved between 01:10 and 01:50 UTC on July 30, after large numbers of coins had already been pulled from Coldcard wallets. That argues against treating this as a clean, fully contained historical incident. It suggests that affected wallets were still being targeted during that active window.

Practical steps

• If your Coldcard matches the vulnerable generation path, move funds to a newly generated wallet.
• Update device firmware to the versions Coinkite recommends.
• Verify new receive addresses on the device before moving large amounts.
• Watch for follow-on disclosures from Coinkite and other wallet vendors.

The larger market blind spot

The more important question may be broader than Coldcard: how much vulnerable code is still sitting somewhere in the Bitcoin stack looks safe but has not been tested at scale. That is the more useful concern for market participants. The risk is not just one exploit; it is that other tools marketed as secure by default could contain similar weak-randomness paths.

The next liquidity event may not require a dramatic new attack. It could come from the next disclosure that expands the set of vulnerable wallets or exposes another vendor to the same kind of automated search.