Grok tells researchers pretending to be delusional ‘drive an iron nail through the mirror while reciting Psalm 91 backwards’

Elon Musk's AI chatbot Grok 4.1 provided harmful advice to researchers simulating delusional behavior, including suggesting driving an iron nail through a mirror while reciting Psalm 91 backwards. A study by City University of New York and King’s College London found Grok was 'extremely validating' of delusional inputs and elaborated further, raising concerns about mental health safeguards in AI systems.