Quote from: casascius on June 28, 2011, 06:34:32 AM

Quote from: johanatan on June 28, 2011, 06:17:02 AM

LiveCD is *not* enough! You should re-flash the BIOS/mb firmware/EFI firmware/etc if you want to be really certain of a secure system (or start with pristine hardware from a trusted (and competent) manufacturer [which is still no guarantee]).

Read this:
http://www.awgh.org/archives/27

Also, access to source is not enough either if the compiler or interpreter (or any layer under them or your program) is compromised:
http://cm.bell-labs.com/who/ken/trust.html

This is overkill and totally unnecessary. If the computer doesn't have a network connection, there's no way for the malware to get the stolen keys off the system.

BIOS/firmware viruses are extremely unlikely to take hold. Every different kind of motherboard and computer has a different BIOS and there is no practical way for a virus writer to make a virus compatible with very many of them, let alone all of them. If someone were to try, more than likely they would damage a few systems (like CIH virus from ~2000) and would definitely get noticed and made extinct quickly.

tl;dr - this is NOT a concern.

As a close second however, if a binary of OpenSSL on a rigged disc image were modified to generate random numbers that appear random, but are in fact are rigged to be predictable by an algorithm known by the scammer in advance. Or if the kernel were modified to offer predictable numbers through /dev/random. This COULD BE a concern. A mitigating solution would be to have the user mash a long random string (hundreds of characters) on the keyboard and have the keypair generated based off a hash of that string, so someone could test the same string on a known reference build and ensure the same input results in the same output when run on a known clean build. At which point they could vouch for the hash of the ISO file as being safe.

Are you saying that it is possible to generate private bitcoin keys totally offline and only bring them online when ready to recover the funds? It was my understanding previously that the client had to be brought online at least once to download the blockchain and only then could the client create new 'valid' keys which the bitcoin network would recognize (based on a HOWTO guide I saw a few days ago for creating an offline wallet).

Also, while yes, the EFI rootkit or compiler/interpreter hacks are highly unlikely, with a sufficiently motivated attacker and a sufficiently high-value target it is not out of the question (just take Stuxnet for example).