AI researcher Pliny the Liberator bypasses safety guardrails of xAI's new Grok-4.5 model using reframing techniques · Digg