Quote from: klokan on July 20, 2014, 06:26:15 PM

1/ To "dump this device memory" you need expensive laboratory equipment and time. I'm not sure how much time you have, but it should be enough to transfer the BTC if the device is lost/stolen. One more comment on this topic: You should still take care of the physical security yourself. But if you do, then with this device you can safely send/receive BTC what was not possible pre-Trezor, because you used (potentially) unsecure PCs.

Good to know that imaging the device is not as simple as plugging the internal memory (SD Card for example) into a PC and bruteforcing the PIN in seconds. I think that should be stated more clearly in the website. You say that the attacker would need expensive laboratory equipment, can you please be more specific, or point me to the doc where they describe this point? It would be important to discard this attack surface.
In any case it would be a good idea to have another Trezor at home ready to send all the funds to a paper wallet, just in case the main one 'disappears'.

Quote from: klokan on July 20, 2014, 06:26:15 PM

2/ Thezor connects as Human Interface Device (HID) and sends Google protocol buffers over the USB wire. It does not need driver on PC. When it received data over USB, it parses and validates protobufs and invalid data are rejected. Interface does not allow private keys to leave the device so any valid reqest that is answered by the device cannot leak the keys. The worst what "malitious driver" can do is getting your public keys and btc adresses and it can send various requests to the device that will prompt you for pin entry and confirmation. These include: signing transactions or wiping the device.

If the first point is solved, and considering that the integrity of the device can be verified by providing a test seed and checking the generated addresses, the only attack surface I can think of is the validation of the data sent by the computer to the device. I'll take a look at how Google's protobuf works.

Is there any reason why they went with BIP 0032 instead of simply SHA256(SEED+idx) or similar? The added complexity of BIP 0032 is justified when we need 'watch only' wallets for example, but in the use case of Trezor, where there is a single user and the privates keys are not supposed to be leaked I don't understand the reason. Maybe they plan to allow Trezor to only receive payments, but not send them? (for example, for waiters in a restaurant).

I find some points unclear in their FAQ. For example, how does the passphrase work? They only say that it sits "on top of the PIN". So am I supposed to write it in the computer's keyboard? Also, in their sentence "[...] we have implemented a safe way of entering PIN, so no keylogger can be used to spy on your PIN" there is no information. I finally found in reddit a video showing how the PIN input works (but not yet the passphrase).