You know that current AI systems are not reliable and produce errors?

How do you protect your life and the life of others using your software against potential lethal errors?