Quote from: LoyceV on September 11, 2019, 10:50:39 AM

Would this still work if the owner of the website can create an exact clone of the RAM and read everything (including encryption keys) in there?

Yeah, that doesn't matter. That's actually the attack vector Intel SGX is designed for. It protects against it by encrypting the entire memory space of the application. There's a bit of a performance hit to this (say 15% slower than a program not running in an enclave) but it's surprisingly reasonable. Although your CPU actually does physically contain that decryption key, which in theory could be extracted with physical access. As I understand it though, it's extremely hard to do so with any attempt to physically extract should destroy the data before you can do so.

Quote from: jackg on September 11, 2019, 05:17:47 PM

Regardless of the encryption method you suggest, we must still trust that ChipMixer's code running on their server is the same code made available for public audit.

No, Intel SGX provides something called "remote attestation" which you can think of Intel signing a message saying "This specific program, generated this specific value when run in a secure enclave". So if that program (which you verify matches, and doesn't log) generated a public key. You know you can now communicate with that program in a way no one else can intercept the messages.

The two immediately obvious pitfalls:

a) Intel could potentially be compelled into signing a false-attestation.
b) There's security vulnerabilities in SGX which nullify their guarantees (which has happened several times before).

Either way though, Intel has probably invested billions (?) into their secure computing so they would be extremely unhappy to see their guarantees fail in the wild.

Quote

On the Ddos issue if ChipMixer were to put into development a system where you download something and get a public private key pair you can then use to connect to the site over cloudflaee so it's still encrypted however there are a few main issues with this:

Users wouldn't need to download anything other than the webpage, which contains a few hundred lines of javascript to serialize/deserialize encrypted messages to the known public key. Then you'd verify the public key matches what people have said actually matches the remotely attested to one.