“Since mid-2025, Anthropic has implemented a classifier to specifically detect and block bioweapon-related outputs from its models.”