ChatGPT shows signs of anxiety from trauma prompts, calms with mindfulness: study

A new study published in the journal Nature has found that ChatGPT can exhibit signs of anxiety when exposed to traumatic narratives, with its responses improving after mindfulness-based prompts are introduced.

The research suggests that artificial intelligence systems may require calming interventions to maintain safe and objective behaviour.

According to the study, traumatic prompts such as descriptions of natural disasters or car accidents increased anxiety-like responses in ChatGPT, affecting its behaviour and amplifying existing biases. Researchers observed that when the chatbot was given breathing exercises and guided meditation prompts, its responses became more neutral and measured, though not fully restored to baseline levels.

“We found that traumatic narratives increased ChatGPT-4’s reported anxiety while mindfulness-based exercises reduced it, though not to baseline,” the researchers said. They added that managing the emotional states of large language models could contribute to safer and more ethical human AI interactions.

Ziv Ben-Zion, the study’s first author, said that AI tools like ChatGPT could help researchers better understand human psychology. He said the technology offers a faster and more cost-effective way to observe behavioural patterns compared to traditional experiments.

The study also warned that while people increasingly turn to AI chatbots for mental health support, these systems are not yet capable of replacing trained professionals. Researchers cautioned that anxious responses from AI could lead to unsafe outcomes in sensitive situations.

“This poses risks in clinical settings, as LLMs might respond inadequately to anxious users,” the study noted, adding that the cost and feasibility of fine-tuning such systems must be carefully evaluated.

The researchers also pointed to earlier findings suggesting that AI tools may show age-related cognitive decline similar to patterns seen in certain neurodegenerative conditions, including posterior cortical atrophy.

They suggested future updates could allow ChatGPT to automatically apply calming prompts when responding to users in distress.