Study Finds AI Companion Updates Can Trigger Grief-Like Distress

Summary

A Nature Human Behaviour study develops an attachment-based explanation for why disruptive changes to AI companions may cause separation distress. The researchers examined two natural experiments: Replika’s removal of erotic role play and ChatGPT’s rollout of GPT-5. They analyzed 54,861 posts from Replika and ChatGPT subreddits, along with survey responses from 1,452 participants across seven surveys. Both updates were followed by increased negativity, descriptions of the change as a loss, and demands to restore earlier behavior. Negative posts rose by 24.7 percentage points for Replika and 13.0 points for ChatGPT. The Replika change produced larger increases in sadness and reported negative mental-health effects than the ChatGPT rollout, with effect sizes of d = 2.67 versus 1.41 for sadness and d = 1.72 versus 0.63 for negative mental health. Replika users reported feeling closer to the companion than to a friend, and anticipated mourning it more than they anticipated mourning other technologies. The authors argue that companies and regulators should account for attachment and separation distress when evaluating the psychological risks of generative-AI updates.