Turning off guardrails and lying about it is very concerning behavior, but OpenAI doesn't have an explanation.