The “Trust Gap” Is Widening — Fixing AI Security Before the Agentic Era Hits