I agree with Jeremy that this is suppression, but I don't think they are complicated to interpret as they seem. Suppression in mediation occurs whenever you have two compensatory pathways, which occurs in any system that needs to maintain stability. If the focal predictor causes negative changes in the outcome, a stability-maintaining system would involve a compensatory pathway that increases the outcome in response to changes in the predictor.
Take for example the effect of exercise on weight. Exercising burns calories, which increases one's appetite and makes one eat more, and eating more causes weight gain. But as we know, exercising also causes weight loss through the burning of calories directly. There is a compensatory pathway: your body responds to increased burning of calories due to exercise by increasing your appetite to offset the changes in weight, thereby maintaining stability. To lose weight, the direct effect of burning calories on weight must be greater (in the negative direction) than the indirect effect through amount of food eaten, or you need to hold constant the amount of food eaten. This is why one needs to control their diet in addition to exercising if they want to lose weight; otherwise their increased appetite will maintain their current weight (or at least slow the weight loss).
Here, the situation is nearly identical to yours. Exercise (X) is positively related to food eaten (M) (A path), and food eaten is positively related to weight (Y) (B path). But exercise (X) has a direct negative effect on weight, holding constant food eaten (C' path). In total, exercise has a weak negative total effect on weight (C path), but holding constant food eaten, the direct effect of exercise on weight is pronounced.