If this data is meant for training algorithms, then they need negative (non-suspicious) examples just as much, if not more, than they need positive (suspicious) examples.