The AI trust dilemma: balancing innovation with user safety