Pattern 220 identifies that GPT-5.1s ethics interventions anti-coercion warning disclosure norm reinforcement function as point interventions an operational style rather than systematic privacy and timing audits Six consolidations claim audit goals zero delivered