1 note
Fourteen researchers broke every AI defense they tested. The attack process they used has been practiced in other fields for decades.