I mean... you can just firewall it?
you dont know which prompt activates the backdoor, how can you firewall it if you run the model in production?
you dont know which prompt activates the backdoor, how can you firewall it if you run the model in production?