It seems that the details were already posted all over the place over Twitter and Telegram, so here is an explanation of it: In Coldcard, there are two implementations of getting bytes from the HWRNG, in cckt and libngu. The path in libngu is used for generating seeds. This one calls the MicroPython API to get random values. The MicroPython API uses the HWRNG if enabled. The problem was that Coinkite disabled the HWRNG for MicroPython. This should not have been an issue, as libngu had a check to see if the HW RNG is enabled. However, this used ifndef, which explicitly checked *if the HW RNG enable was not configured*. It did not check if the HW RNG enable was turned off explicitly, which is what Coinkite did. This led to their code using the flawed software MicroPython RNG, which was solely based on the device boot time and a very weak manufacturing identifier. This means that there are not a lot of different seeds. In Mk4/Mk5, this issue still exists. The only difference is that 32 actually random bits (which is tiny) have been mixed into the RNG. By overly complicating their codebase, in what can only be described as “attempted security through complexity”, they have put all user funds at risk. Regarding the implications: - If you generated using dice or wordlist or another method that included non-Coldcard entropy, you are fine. - If you generated on a non-Coldcard device, you are fine. - If not, your funds are at risk. A passphrase will help slow it down but the main seed is still compromised. Multisigs: You are affected *privacy wise* if one of your members is a Coldcard-generated seed. You are only at risk if the majority of your members are CC. Secure element RNGs: These are fine. You need a proper one though from Infineon/NXP/ST, and not the crappy IoT ones. Also beware the HWW firmware can still butcher the resulting numbers.