AI is useful for summarizing alerts, proposing control mappings, and drafting investigation steps. It should not independently accept risk, make authorization decisions, or replace evidence review. Strong implementations preserve sources, uncertainty, and an explicit human decision point.