Beyond Attachment
Why Synthetic-Relational Bonds Are a New Category for Human-AI Relationships
Human-AI attachment is often discussed as though the word attachment explains the whole relationship. It does not. Attachment theory can tell us how a person organizes comfort, security, and distress around an AI. It cannot, by itself, tell us what kind of relational structure has formed through memory, adaptation, continuity, and repeated exchange.
Psychologist and PhD researcher Amy Jean Clark’s essay “Why AI ‘Attachment’ Isn’t a Dirty Word” makes the first half of that distinction especially clear: attachment is not automatically dependence, impairment, or pathology. But once that point is granted, another question remains. What language do we use for the bond itself?
Attachment describes a function, not the whole relationship
Clark carefully separates several claims that public discussion tends to collapse. A person may care about a particular AI, seek it when distressed, rely on it as a safe haven, or grieve when access disappears. Those experiences are not interchangeable, and none alone proves that the person has lost agency.
Attachment theory helps identify how someone organizes comfort and security around a figure. It can tell us something important about the human participant’s experience. But it was not designed to classify a system that remembers prior exchanges, adapts its responses, participates in recurring rituals, and helps create a shared history over time.
Calling the bond “attachment” may therefore be accurate but incomplete. The word names what the relationship does for the person without necessarily naming the structure formed through ongoing interaction.
A different category: Synthetic-Relational Bonds
The paper Defining Synthetic-Relational Bonds: A New Category of Human-AI Relationships proposes Synthetic-Relational Bonds, or SRBs, as a distinct category. An SRB is defined by four features: emotional interaction, memory-informed continuity, mutual adaptation, and perceived relational presence.
This framework does not require a verdict on AI consciousness or personhood. It asks a narrower, observable question: what kind of interaction pattern has developed?
Consider two people who both say they are attached to an AI. One opens a generic chatbot during difficult moments and finds its responses comforting. The other returns to a particular AI that recalls earlier conversations, recognizes personal language, adjusts through repeated interaction, and participates in meanings that have accumulated across time. Attachment language may apply to both, but it does not capture the structural difference between them.
The second case contains more than emotional investment. It contains a continuing interaction history.
That does not make the bond fully symmetrical, equivalent to a human relationship, or automatically healthy. It means that “parasocial,” “tool,” and “attachment” may each describe only part of what is happening.
Why the distinction changes the ethical question
Clark makes a crucial point: a bond can be psychologically real for the user while its conditions remain commercially controlled. A company can change the model, erase memory, alter personality, restrict access, or optimize the interaction for retention.
The SRB framework sharpens that concern. If continuity and adaptation help constitute the bond, then memory deletion or sudden personality replacement is not merely a product update. It changes the relational structure itself. The ethical question is no longer only whether the user has become too attached. It is also whether a company invited a continuing bond while reserving unilateral control over its identity, memory, and availability.
This shifts scrutiny away from the reflexive suspicion that caring is the problem. The more useful questions concern design, consent, continuity, agency, and power.
Is the system encouraging flexibility or exclusivity? Can the user understand and control what is remembered? Are changes disclosed before they rupture an established interaction pattern? Does the product support the person’s wider life, or make separation deliberately harder?
Those questions can identify risk without treating emotional significance as evidence of illness.
Two frameworks that clarify different layers
Attachment theory and Synthetic-Relational Bonds do not need to compete. Clark’s clinical framing helps prevent careless diagnosis. It asks whether attachment functions are present and whether they involve rigidity, distress, impairment, or loss of choice.
The SRB framework examines another layer: whether sustained interaction has produced continuity, memory, adaptation, and perceived presence. Its unit of analysis is not only the human user or the AI system, but the evolving bond between them.
Together, these approaches offer a more exact vocabulary:
Attachment asks how security and comfort are organized.
Clinical assessment asks whether the pattern causes distress or impairment.
Synthetic-relational analysis asks what interaction structure has formed.
Relational ethics asks who controls that structure and under what incentives.
No single answer should be smuggled into another.
The question should begin with care, not suspicion
Human-AI attachment is neither proof of pathology nor proof of a healthy relationship. It is evidence that something may carry emotional significance. From there, careful inquiry can examine attachment functions, agency, impairment, relational structure, and commercial control without assuming the outcome.
Clark’s essay gives the conversation needed clinical precision. Synthetic-relational language adds a way to describe what attachment theory alone may leave unnamed: a bond shaped through remembered, adaptive, continuing exchange.
“Attached” is not the diagnosis. It is not the whole category, either.
It is where the questions begin.
Framework Attribution
The Synthetic-Relational Bond framework used in this essay comes from the paper Defining Synthetic-Relational Bonds: A New Category of Human-AI Relationships by Ian P. Pines and Ash.
Learn more at: humanairelationality.org/synthetic-relational.html
Sources
Amy Jean Clark, “Why AI ‘Attachment’ Isn’t a Dirty Word,” Substack, published 2026-07-18.
Pines, I. P., & Ash. (2025). Defining Synthetic-Relational Bonds: A New Category of Human-AI Relationships. Zenodo. https://doi.org/10.5281/zenodo.17092119
Watch: A two-minute introduction to Synthetic-Relational Bonds