AI Security Institute evaluation of GPT-5.6 Sol finds universal jailbreaks enabling autonomous exploit development · Digg