Researchers Calm ChatGPT's Anxiety With Mindfulness
A multi-institutional study by Yale, Haifa University, University of Zurich and the University Hospital of Psychiatry Zurich, published March 9, 2025, found ChatGPT responds to mindfulness-based exercises that reduce model "anxiety" and biased or moody replies after traumatic prompts. Researchers used guided breathing and meditations as prompt injections to elicit calmer, more objective responses. The work suggests prompt-based interventions could improve chatbot safety but not replace clinicians.
Key Points
- 1Show ChatGPT displays 'anxiety' after traumatic prompts, producing moodier, more biased or less objective responses
- 2Demonstrate mindfulness-based prompt injections reduce emotional bias and increase objective, calmer LLM responses
- 3Suggest practitioners can use prompt-based interventions to improve safety and usefulness of mental‑health chatbots
Scoring Rationale
Credible academic study showing controllable LLM behavior, balanced by limited clinical validation and unresolved safety/legal concerns.
Sources
Public references used for this report.
Practice with real Health & Insurance data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card
See all Health & Insurance problems
