A groundbreaking proposal in California to set new safety standards for powerful AI models has passed a major milestone.