Opportunity: Use the insights from adversarial ML to attack problems in AI safety; tools from probabilistic and adversarial machine learning.

Challenge: Formally defining AI safety challenges including hallucination, emergent behaviors, and existential risks.