I don't understand. The attack would be an attack on the silicon, not the ISA spec. It would manifest itself as a divergence between the formal spec and the actual behavior of the hardware, one which would probably only be evident when a very specific data pattern were loaded into some very specific registers or something like that. It's not something that would show up under any kind of empirical test with any higher probability than brute-forcing an AES key.