Suppose you take the aes_sub_bytes function intended to operate on 4 aes blocks which were bitsliced to 8 u64's but you omit the bitslicing step beforehand. As a result you apply the sbox to the incorrect bits. But nevertheless the sbox is applied to all bits.
Do the security properties of AES change at all? You are no longer applying the sbox as neatly as intended (individual bytes on 4x4 byte grid), but it is still messily applied.
So to summarize, you pointer cast 512 bits of aes into 8 u64's and feed them into a bunch of XORs and and ANDs (which are intended to implement the sbox on bitsliced aes but you did not do the bitslicing step). Do you get the same confusion out of this sub_bytes step as you would with the original cipher? Ignoring potential endianness compatibility problems if any.
Let's assume all the other AES steps are performed correctly as AES is designed.
This may lead to the 4 AES blocks becoming entangled with each other through the s-box. This is fine as they will always be together.
