> You can actually discover those in open weight artifacts, reproduce them, study them and issue a security bulletin. Finding unknown backdoors in models is NP

> You can actually discover those in open weight artifacts, reproduce them, study them and issue a security bulletin. Finding unknown backdoors in models is NP hard.