Unbelievable what has happened. I own both a Coldcard Mk4 and a Q. I appear to be unaffected by this specific flaw because I used sufficient dice rolls to generate my seed, but I feel terrible for the victims.
What I cannot understand is how nobody discovered this earlier. On the affected Mk3 firmware, seeds were reportedly generated with only about 40 bits of entropy. On the Mk4, Mk5 and Q, entropy could be as low as approximately 72 bits.
Seed generation is arguably the single most important part of a hardware wallet’s source code. How could such a fundamental flaw remain undetected for years and then be carried across several generations of devices?
I am waiting to hear Coinkite’s full explanation—not only how the error happened, but why its internal review, testing and external scrutiny failed to identify it. I also want to see how the company responds to the victims.
Even though my own wallet appears unaffected, I am seriously considering abandoning Coldcard. My decision will depend heavily on the quality and transparency of Coinkite’s response, and on what it does for those who lost their funds.
Well After reading this thread I have decided to scale back my BTC holdings. If it is a big deal and if AI routinely attacks and beats safety measures as this was obvious AI crypto will tank like mad.
I do not need too much exposure in the space. I sold about 30% of my coins today.
wth, do you really think AI can crack private key? This Coldcard issue or any of the recent exploits have nothing to do with AI "breaking safety measure". In most cases, these are caused by our errors. Maybe supply chain attack or firmware vulnerabilities or bad implementation.
AI is just a tool, I dont think it can brute-force an ECDSA private key, at least with current technology.
No, I do not think AI “cracked” Bitcoin private keys or broke ECDSA. A properly generated 256-bit private key remains computationally out of reach.
The possible role of AI is completely different. AI could have helped an attacker review the publicly available firmware, trace the seed-generation code, identify that the wallet was producing seeds from a drastically reduced entropy space, and then assist in writing or optimizing the tools needed to exploit that flaw.
In this case, the attacker would not be searching the full 256-bit private-key space. They would be reproducing the much smaller set of seeds that the defective firmware could generate—about 40 bits in the worst Mk3 case and potentially around 72 bits on later models—then deriving standard Bitcoin address paths, checking those addresses against the blockchain, ranking the funded wallets and automatically creating sweep transactions.
AI could also help with code analysis, firmware emulation, debugging, parallelization, derivation-path scanning and automation. None of that means AI defeated Bitcoin cryptography. It means AI may have helped discover and exploit a serious implementation error created by humans.
There is currently no proof that AI was actually used, so that part remains speculation. But saying AI could have played a role is not the same as claiming it brute-forced an ECDSA private key.
