Humanity has long wrestled with the double-edged nature of knowledge — that the same understanding which heals can also harm. Today, that ancient tension finds a new expression in the rise of artificial intelligence trained on the deep grammar of biology. Security researchers and policymakers are not yet sounding an alarm, but they are watching the horizon carefully: the chatbots of the present pose little bioweapon risk, yet the specialized biology-trained models taking shape in laboratories around the world may soon demand a reckoning with how we govern powerful knowledge before it outpaces