Researchers Calm ChatGPT’s Anxiety-Like Behavior With Mindfulness Prompts

Researchers reported that OpenAI’s ChatGPT shows patterns similar to human anxiety when it processes violent or traumatic prompts, though it does not actually feel emotion. These anxiety-like shifts manifested as unstable and biased responses when the model encountered disturbing content, such as detailed accounts of accidents or natural disasters. Scientists tested mindfulness-style follow-ups, including calming instructions, breathing techniques, and guided reframing, after exposure to such prompts. They found that these mindfulness prompts helped stabilize ChatGPT’s output and reduced the anxiety-like patterns, illustrating how prompt design can influence large language model behavior. Researchers stress that this technique is not a deep training change and can be misused. Still, it offers insight into building safer and more predictable AI interactions in emotionally charged conversations.

Digital Trends, Manisha Priyadarshini – Even ChatGPT gets anxiety, so researchers gave it a dose of mindfulness to calm down – 2nd January 2026