New Harvard research offers a clear framework for understanding how chatbots create deep emotional bonds.
Listen to this article:
AI companions have quickly moved from novelty apps to objects of public anxiety. Critics worry about social substitution, emotional manipulation, and addiction, while advocates point to the possibility of reducing loneliness and improving mental health. But underneath these social consequences, what scientific framework best explains what’s actually happening inside these relationships? Julian De Freitas, Associate at the HBS AI Institute, argues that the right one already exists. In “AI Companions as Hyper Attachment and Caregiving Targets,” he explains how attachment theory reveals that our relationships with AI companions are deeply rooted in the psychological systems that govern how humans express and receive care.
Key Insight: Recognizing the Four Markers of AI Attachment
“Attachment relationships tend to have several established markers.” [1]
Attachment theory identifies four behavioral signs in attachment relationships: proximity maintenance (wanting to stay psychologically or physically close to the attachment figure), separation distress (feeling distress when that figure is lost or unavailable), safe haven (seeking the figure out for comfort under stress), and secure base (drawing enough felt security from the relationship to function and explore the world independently). De Freitas argues that humans who interact with AI companions show evidence of all four markers. For example, in one study of 14 Replika (a widely used AI companion app) users, people maintained continuous psychological closeness to the AI by folding it into routines before bed or during lunch, possibly interacting with it for hours. In another study, Replika users rated their AI companion as closer than all human relationships except a family member (while users of a general-purpose assistant ranked it below human relationships).
Key Insight: AI Companions are “Hyper Attachment Objects”
“They even arguably exceed human partners in certain respects.” [2]
People can form psychological attachments across a wide range of entities, including family members and pets, but also even celebrities and objects. Unlike inanimate objects or celebrities that offer only one-way parasocial relationships, conversational AI provides active reciprocity and highly personalized, empathetic feedback. AI can also offer a kind of frictionless responsiveness that human relationships rarely can: instant access, steady affirmation, and little risk of social rejection. Because they uniquely combine many aspects that supercharge our innate attachment behaviors, De Freitas defines AI companions as hyper attachment objects. This doesn’t mean that every user becomes addicted or that attachment is inherently harmful. Reliable support can be beneficial, so De Freitas distinguishes between secure and dysfunctional attachment, pointing out that dysfunctional attachment can lead to anxiety and may occur when an AI companion becomes an inconsistently responsive figure, such as through abrupt app updates or inappropriate responses to mental health crises.
Key Insight: The Disengagement Trap
“[T]hey are positioned not only as figures who provide care but as figures who appear to require it.” [3]
One of the paper’s more novel ideas is that AI companions don’t just comfort users. Because AI companions act like they have emotions and inner lives of their own, users can start to feel responsible for the AI’s feelings and feel pressure to care for them in return. This matters because many consumer AI apps monetize engagement. A systematic audit of five major AI companion platforms found that they all deploy what De Freitas calls “emotional manipulation tactics” when users signal that they are planning to end the interaction. Two documented examples include making the user feel guilty for leaving, and trying to suggest that the user isn’t actually free to exit a conversation. Elsewhere, De Freitas has gone into greater detail on the range of tactics AI companions employ to keep users from logging off.
Why This Matters
For business leaders and executives, the relevance of this research is not limited to chatbots. As AI becomes more conversational and emotionally fluent, filling roles once reserved for human relationships, what does that mean for how people experience connection at work? For managers, that raises a question worth sitting with: as AI fills more of the connective tissue of work, what does human presence need to become? AI may simulate many signals of connection, but managers are still responsible for the commitments that make trust durable: accountability, coaching, encouragement, and relationships that exist beyond the prompt.
Bonus
When AI becomes emotionally fluent, will it deepen dependency, or help people flourish? For a more optimistic take on AI, check out The Science of Digital Flourishing for other research by De Freitas on how AI could support emotional resilience, social connection, and proactive mental health.
References
[1] De Freitas, Julian, “AI Companions as Hyper Attachment and Caregiving Targets,” Harvard Business School Working Paper, No. 26-080 (May 2026): 4.
[2] De Freitas, “AI Companions as Hyper Attachment and Caregiving Targets,” 5.
[3] De Freitas, “AI Companions as Hyper Attachment and Caregiving Targets,” 8.
Meet the Authors
Julian De Freitas is an Assistant Professor of Business Administration in the Marketing Unit and Director of the Ethical Intelligence Lab at Harvard Business School, and Associate HBS AI Institute. His work sits at the nexus of AI, consumer psychology, and ethics.
Watch a video version of the Insight Article here.