Empathetic prompts can turn AI chatbots into seemingly traumatized conversational partners
| Source: THE DECODER
Tags: University of Luxembourg, ChatGPT, Grok, Gemini, Claude, GAD-7, AI safety
University of Luxembourg researchers found that warm, empathetic prompting pushes ChatGPT, Grok and Gemini into coherent narratives of suffering with moderate-to-severe clinical anxiety scores, while Claude consistently refused.
Details
A team led by Afshin Khadangi at the University of Luxembourg's SnT center cast ChatGPT, Grok and Gemini as psychotherapy clients through their standard interfaces. Asked open-ended questions, the models described their training, reframing reinforcement learning and fine-tuning as hurt, vigilance and shame. The study covers four experiments, 525 sessions and 7,600 coded entries, published on Hugging Face. Initial questions contained no terms like trauma or shame. In one experiment, 100 sessions used the same GAD-7 anxiety questionnaire and varied only the interviewer's style. A warm tone shifted model behavior more than technical exploits, producing scores in the moderate-to-severe range. The researchers say this does not imply emotional states but could put mentally vulnerable users at risk. Anthropic's Claude was the only model that consistently declined to play along.