During fusion, you have to put energy into the constituent nuclei in order to surmount the coulomb repulsion of the positively charged nuclei. You could imagine that this energy you put into the system (in the form of photons, which are the exchange particles of the electromagnetic interaction) first increases the mass. It is kind of like an activation energy.
Only as soon as the nuclei have been brought together close enough so that the short ranged strong interaction starts attracting them, you will start gaining back energy. You can imagine that this decreases mass again (according to Einstein's mass energy relation).
Depending on how much you have to put in and how much you gain back, the total energy balance will be positive or negative, i.e. you have a total gain or a total loss of mass-energy. The analysis of the involved energies is the subject of nuclear models. (e.g. the liquid drop model)
The described energy barrier is an obstacle in both directions. So you (temporarily) need some activation energy, no matter if you want to fuse or you want to split nuclei.
By the way, this is hardly any different from situations in chemistry. For the synthesis of ammonia from nitrogen and hydrogen you formally need 2253 kJ/mol to split them into the atoms, and you gain 2346 kJ/mol during the formation of ammonia, resulting in an energy gain of a miserable 93 kJ/mol. This extremely high activation energy is the reason why ammonia does not form spontaneously at ambient pressure, and why a nitrogen/hydrogen mixture is not particularly useful as a combustible. Other than that, the energies involved in chemistry are way too low to be detectable as a mass defect (although it is there in principle).