AI agents in security can now investigate, decide, and act before an analyst opens a ticket. A CertiK report argues that once they can, companies need defined authority, human handoff points, decision records, red-teaming, and named accountability, . ...