Will transparency help catch deception? Perhaps not — AI Alignment Forum