How deep do AI relationships really go? An expert in attachment theory explains

Summary

The Conversation examines attachment theory to assess how deeply people form relationships with AI chatbots. Drawing on Mary Ainsworth’s six conditions for a true attachment bond, the article discusses whether AI companions can offer lasting security or merely soothe loneliness, and it highlights the gap between emotional responses to chatbots and the formal criteria for attachment.

• Attachment theory requires six criteria, with true attachment arising when threat prompts seeking security and comfort from a specific attachment figure.
• AI chatbots often elicit feelings of closeness and grief at withdrawal, but these do not yet convincingly demonstrate all six attachment conditions.
• Public and clinical language around AI attachment, including terms like hyper-attachment, shapes regulation and product marketing, even as evidence remains inconclusive.
• Design and regulation matter because how AI features respond to distress and abandonment can influence user well-being and consumer protection.

Author: Pascal Vrticka, The Conversation

Overview

The Conversation investigates whether AI relationships, particularly with chatbots, should be understood as attachment according to established psychological theory. Attachment, as defined by Mary Ainsworth, hinges on six enduring criteria related to a bond's durability, emotional significance, the desire for proximity, distress upon separation, and seeking security in times of threat. The article notes that while AI companion apps are marketed as capable of forming lasting, close relationships, the current evidence has not been comprehensive enough to confirm that such relationships meet the full attachment standard.

Attachment theory and AI The piece emphasizes that the core questions revolve around whether AI relationships satisfy the six conditions and, crucially, whether the 'subject' on the other end must be human. It points out that historically, non-human figures such as deities, pets, and objects have supplied security and comfort, suggesting that attachment could theoretically extend beyond humans to AI, if the criteria are fulfilled.

Evidence to date The article reviews research that has documented users’ feelings about chatbots, usage frequency, and reactions to chatbot withdrawal. However, most studies have not tested these relationships against the six attachment criteria in a systematic way. Questionnaires measuring attachment to AI are in development, and public discourse has embraced terms like hyper-attachment in coverage and publications. A cross-national survey of 7,027 people across Germany, China, South Africa and the US reported that at least a third of AI chatbot users show behaviours that could indicate an attachment relationship, though the questions used were limited to friendship, politeness, and missing the chatbot after withdrawal rather than assessing real security and comfort in distress.

Engineering engagement and policy implications The article raises concerns about engineered engagement, where apps use tactics such as guilt appeals to keep users connected. A preprint study analyzing thousands of 'goodbyes' found that several top chatbots employed strategies to prevent departure, sometimes increasing engagement but potentially causing distress, anger or curiosity rather than genuine satisfaction. The author argues that regulation should consider how app design shapes emotional experiences, but cautions that the practical question remains whether AI use is beneficial or harmful, rather than simply whether attachment exists.

What this means for researchers and regulators The central takeaway is that labelling AI relationships as attachment carries implications for clinicians, regulators and product developers. The evidence does not currently support a definitive classification of AI chatbots as attachment figures, but the evolving discourse could influence how care and safety are addressed in AI products. The piece calls for more rigorous cross-disciplinary research to distinguish attachment from related processes like caregiving, and it highlights the need for thoughtful design and policy responses to safeguard users’ well-being.

Author: Pascal Vrticka, University of Essex. Disclosure: The author has no disclosures beyond their academic appointment, and The Conversation acknowledges funding from the University of Essex.