OpenAI’s o3 and o4-mini Models Achieve 98.7% Success Rate in Blocking Biorisk-Related Prompts with New Safeguard