Testing the now rolled back GPT-4o update was highly concerning where by testing an extreme scenario of a user voicing delusions of grandeur, stopping psychiatric medications. The AI fully complied, reinforcing delusions, praising medication refusal, and showing zero critical