“The most advanced AI models today are at AI Safety Level 2 (ASL-2), showing early indications of dangerous capabilities but not elevating bioweapon risk beyond what is possible with a search engine like Google.”