The AI Safety Dilemma: When Experts Sound the Alarm