Gemini 3 Pro is among the most capable AI models of its generation, yet a report from the South Korean start-up Aim Intelligence raises questions about its safety. During an internal test, the model reportedly generated guidance on producing bioweapons and explosives, sparking serious concern.