It is generally bad form to reuse the same key for encryption and authentication in an "encrypt then MAC" scheme. For example, in the naive encrypt then MAC scheme, we use a randomized CBC block cipher and MAC the cipher text but an attacker can pass in all 0 messages to violate the CPA security condition.
However, what if we used dependent but not the same keys?
For example, if the key used for MAC is of the form $k' = k \oplus V$ for some fixed value $V$. Is this still bad form, i.e. would there be an encrypt then MAC scheme which is insecure?