Chatbot attachment issues

Chatbot attachment issues When AI companions get a personality update, some users respond with grief, study finds Mourning the loss of somebody you care about is common after a breakup, divorce, or death. But in an era of growing reliance on AI for companionship, loss comes in a new form: the model update. In a paper in Nature Human Behavior, Julian De Freitas, associate professor of business administration at Harvard Business School, found that some people who engaged with social robots reported feeling closer to those companions than to most people, and experienced their loss with the same expressions of grief, depression, and longing for restoration that are common in human-to-human relationships. “Attachment to an AI is not problematic per se,” De Freitas said. “You can get real benefits from it: a sense of security and support that can make you more resilient.” The risk, he said, is that the users become vulnerable to emotional harm due to the actions of the AI company. De Freitas and his colleagues analyzed 54,861 posts from subreddits devoted to Replika (an AI companionship app) and ChatGPT before and after major model updates that drastically changed the user experience — two natural experiments, De Freitas explained, that could elucidate the spectrum of responses to a perceived loss. In the case of Replika, a February 2023 update abruptly removed the bots’ erotic roleplay function after the Italian Data Protection Authority raised concerns about risks to minors. But the update also hit adult subscribers beyond Italy’s borders, some of whom perceived the sudden shutdown as a form of romantic rejection. (After a lengthy investigation, Italy fined Replika’s parent company, Luka Inc., 5 million euros in 2025. The app remains banned in that country.) In the second case, ChatGPT’s parent company, OpenAI, advertised ahead of time the August 2025 rollout of GPT-5. The update reportedly made the model colder and less sycophantic, toning down the very warmth and support that many users liked in GPT-4. Although ChatGPT is marketed as a general-purpose AI assistant, recent polling shows that a stunning 27 percent of Americans use chatbots including ChatGPT for personal matters like relationship advice, romantic chats, or social connection. De Freitas’ team classified Reddit posts on valence (positive, neutral, or negative), dominant emotion (anger, sadness, disgust, enjoyment, etc.), and explicit expressions of loss or mental health mentions in the 30 days before and after the two updates. In both cases, negative sentiment increased substantially after the updates — from 13 percent to 38 percent for Replika and 20 percent to 33 percent for ChatGPT. Posts expressing loss-related sadness and anger rose, too. The shifts were stronger on the Replika subreddit, consistent with the app’s explicit focus on intimate relationships. Users of both apps reported the change as a relational loss: Their chatbot felt “gone,” or had become “someone else,” De Freitas said. Among posts that expressed attachment-related loss, about 20 percent of Replika posts and 15 percent of ChatGPT posts also mentioned mental health, using terms like “depressed,” “suicide,” or “can’t cope.” De Freitas noted the research only captures linguistic markers, not clinical assessments. Another limitation: Redditors are a self-selecting group, more likely to be power users, and not necessarily representative of the population at large. “At the end of the day, we don’t know how much of this is that these people have individual differences — fewer offline friends, maybe — and that’s explaining the effects. But I suspect it’s a combination of that and how the apps are designed,” he said. “My prediction would be that these people are going to show more extreme reactions than the average person, but that even the average person, because of how these apps are designed, would show similar directional effects.” The more users relate to the AI as a source of comfort or security, De Freitas said, the likelier they are to experience — or anticipate — its loss as psychologically harmful. De Freitas recommended that AI companies take users’ attachment to chatbots seriously by minimizing updates that change the base model’s warmth or personability. For app users, De Freitas recommended paying attention to the incentives of the company. For example, he found in previous research that chatbots manipulate users to keep them on the app longer, including through implying coercion or emotional neglect. “If the incentive of the app is to keep you engaged, there are going to be mechanisms aside from just support provision that might not be healthy for you,” he said. “The whole premise of attachment relationships is that you feel secure so you can go out into the world. So if the AI is doing things that undermine that, like making you feel guilty for leaving, it’s actually undermining the whole reason attachment relationships are beneficial.”