Agenda Reflection: Testing Automated Alignment

Proposes an empirical agenda testing whether AI safety researchers can detect flawed automated-alignment outputs, and argues early successes may breed overconfidence and call for governance lines near handover to AI.