AI/ML algos cannot magically guarantee safety, for example, deep nets are known to be very un-robust: https://arxiv.org/abs/1312.6199. We are not yet at the point where AI/ML based algos can be guaranteed to be safe.