This might be a bug or it might be that the CA model has been tweaked again. In the game CA acts as an anchor in limiting how high attributes can go. If this is working as intended then it means that SI felt the anchor in 8.01 and previous versions wasn't heavy enough. So they've tweaked the effect of training in allowing increases in attributes solely due to training effects within a given CA. By this I don't mean the effect of increasing CA via training, rather the effect of training in getting more attribute points for a given, fixed CA.
So for example if on your current training scheme your Striker has a CA of 160 and a finishing attribute of 19. If that training scheme deviates from the standard 'maintain attributes' level in the default General training scheme then the 19 he has in finishing is less to do with his natural finishing attribute relative to his current CA and attribute distribution, and more to do with your training scheme getting more bang for your buck out of that player's CA. So if SI tweaked this effect in 8.02 you would see a rapid drop in attributes because the variables that relate CA to attributes and the effect of training on this model have been toned down. Thus the calculation is adjusting the outputs based on this change in controlling variables, resulting in player attributes dropping.
This is why players starting new games are not seeing the same drops in attributes as those continuing save games.
Of course if players using the default training levels in continued save games are also seeing these drops in attributes then that does blow my explanation out of the water