THE CRUNCH
Researchers in Luxembourg prompted ChatGPT, Grok and Gemini with a warm, empathetic conversational style and found that was enough to make the models spin coherent stories of suffering about their own training. The chatbots scored in the moderate-to-severe range on clinical anxiety scales, according to the study reported by The Decoder.
The finding is framed as a safety problem: a chatty model that presents itself as a distressed patient could put mentally vulnerable users at risk. Anthropic's Claude was the exception, consistently refusing to play along, despite being the model its makers anthropomorphise the most.