What definitions or equivalencies between definitions for standard set theory objects (such as large cardinals) do not hold or do not carry through in the expected manner to the world without choice? Also, equivalences that DO carry over, but that you might expect wouldn't, would be interesting as well.
For instance I recently heard that the equivalence between the diagonal intersection and the regressive function definitions of normal ultrafilter does not hold without AC, though under closer inspection I don't see why it doesn't. Clarification on this would be helpful as well.