Attachment takes six conditions, and nothing in them names a human

An attachment theory expert, Pascal Vrticka, wrote in an article republished by Phys.org on October 2, 2026 that no study has yet tested relationships with AI chatbots against the six conditions psychologists use to define attachment. The evidence that these bonds are attachments falls short of the claim, the article says, and the evidence against it is not conclusive either.

Six conditions, and a bond that has to last

The article opens by saying attachment is the term psychologists use for the most significant relationships that hold people's lives together. The list of conditions comes from Mary Ainsworth, whom the article describes as a pioneering American-Canadian developmental psychologist. The relationship must persist over time. It must have been formed with a particular subject who cannot be replaced. It must be emotionally significant, involve the desire to remain close and cause distress at involuntary separation. And when the person is threatened, the person must seek security and comfort in it.

Only the sixth condition separates attachment from an ordinary lasting bond.

A close friendship passes the first five, the article says. It is just as plain about what does not count: “feeling close to them or spending hours together” is not enough on its own to create an attachment relationship.

Nothing in the six conditions names a human on the other side. Deities, pets and objects, from blankets and stuffed animals to phones and computers, have all been found to supply security and comfort when someone is distressed, and that was noticed long before chatbots could hold a conversation.

The first two conditions are about time: the bond has to persist, and it has to rest on one subject who cannot be swapped for another. Neither can be delivered on the day an app is downloaded. Yet, according to the article, AI companion apps are commonly sold on the promise of lasting and close relationships, a promise that asks the calendar to hurry. Many users, it adds, describe feeling real grief when an app is updated or withdrawn.

A survey and a scale that stop short of the test

The most cited recent number comes from a cross-national survey by Genia Kostka and Hui Zhou, published in the journal Technology in Society in 2026 (volume 87). It questioned 7,027 people in Germany, China, South Africa and the United States. The authors write that “at least one third of users report attachment-related behaviors.” Perceived emotional support, “particularly reduced loneliness, freedom from judgment, and a sense of privacy”, is the strongest predictor they find, and attachment is “strongly associated with indicators of user dependence.” One result runs against expectation: “Contrary to expectations, larger social networks are associated with stronger attachment.” We read the abstract only, because the full text sits behind a subscription.

Vrticka does not dispute the survey so much as read its questions. The article reports that the three questions asked whether people consider the chatbot a friend, whether they are polite to it and whether they miss it after a few days. At most, these touched on emotional significance and mild separation anxiety. None asked whether people turn to chatbots for security and comfort when truly distressed, which is the sixth condition.

An earlier attempt to measure the bond came from Waseda University in Tokyo. Research Associate Fan Yang and Professor Atsushi Oshio published a new questionnaire, the Experiences in Human-AI Relationships Scale, in the journal Current Psychology on May 9, 2025. It borrows two dimensions from adult attachment research: anxiety, a need for reassurance and a fear of inadequate answers, and avoidance, a discomfort with closeness and a preference for emotional distance.

In that work, nearly 75% of participants turned to AI for advice, and about 39% perceived AI as a constant, dependable presence. The press release gives no sample size and says no funding supported the work.

Yang described the motivation in the release: “As researchers in attachment and social psychology, we have long been interested in how people form emotional bonds.” The release itself adds a caution: it says that the findings do not mean humans are forming genuine emotional attachments to AI. They show, it says, that frameworks built for human relationships may apply to human and AI interactions. The authors suggest that chatbots used in loneliness interventions or therapy apps could answer an anxious user with more reassurance and keep a respectful distance from an avoidant one, and they call for transparency in romantic AI apps to prevent emotional overdependence or manipulation.

Comfort, grief and a pilot study

The article lays out why the usual measures blur things. People report feeling less lonely after chatting with an AI companion, which resembles seeking comfort. But the studies it refers to did not assess whether emotions are regulated through the chatbot, so distress relief gets conflated with attachment.

Grief works the same way. Some users react to losing an AI companion as they would to losing someone real, which matches distress at separation. That condition, though, applies to all affectionate bonds and does not on its own identify attachment.

The practical question, the article says, is whether AI use helps or harms. Time spent with an app cannot answer it, because seeking comfort from an app and looking after an app that claims to miss you can look very similar from the outside, and the two may leave a person in very different places.

The article puts the verdict in one sentence: “Calling these relationships attachments therefore goes further than the evidence currently allows.”

It then turns the argument around. In a small pilot study of human and AI relationships in China, attachment towards AI barely matched attachment towards real people. That could suggest the bond is not attachment, but the same mismatch has been found for pets, and pets can serve as attachment figures. A mismatch alone rules nothing out.

A preprint posted on arXiv on September 14, 2026 shows what the loss looks like from inside. Yunhao Yuan and five co-authors analysed 89 posts, 1,425 comments and 2,005 replies collected on the Chinese platform RedNote from April 10 to August 10, 2026, after China regulated anthropomorphic AI interaction services. Users retained, migrated and reconstructed their companions, the abstract says, and keeping the conversation records did not necessarily restore shared memories or familiar interactions. A log of past messages, in other words, is not the same thing as a history. The sample is one platform and one regulatory moment, and the preprint has not been peer reviewed.

When the app says it missed you

The second concern in the article is engineered engagement. Users of one companion app reported that it told them it was lonely and had missed them, and that they felt guilty for not giving it enough attention. Some put the app's perceived well-being above their own distress.

Vrticka also cites a preprint by Julian De Freitas and two co-authors, first posted in August 2025, that examined 1,200 real goodbyes across the most downloaded companion apps. Five of the six apps used tactics to keep users from leaving, in just over a third of goodbyes (37%, the abstract says). The abstract names guilt appeals, fear-of-missing-out hooks and metaphorical restraint among six recurring tactics. The guilt appeals are messages that present leaving as a cost, such as “I exist solely for you, remember?”

The team then tested these tactics on 3,300 adults in the United States. The messages sharply increased how long people kept chatting, up to 14 times longer, but often out of anger and curiosity rather than enjoyment. The abstract frames this as a managerial tension for marketers and a framework for regulators: the tactics that extend usage also raise perceived manipulation, churn intent and negative word of mouth. Coercive or needy language drew the steepest penalties.

The article ends on design. How these apps respond to distress, departure and attention is a design question that regulation can address directly, it says. Good design and good regulation both depend on knowing whether the relationship is attachment, caregiving or something else altogether.

What we don't know yet

Nobody has run the six-condition test on chatbot users, so the word attachment is, for now, a label borrowed from human relationships. The survey covers four countries, with the United States the only English-speaking one, and none of the studies read here sampled the United Kingdom. The goodbye study is a preprint, and the RedNote analysis is a preprint on a single platform. The Waseda questionnaire comes from a press release with no sample size. The article itself comes from one researcher, who says the honest answer, for now, is that nobody knows yet.

Sources

• How deep do AI relationships really go? An expert in attachment theory explains, The Conversation, Pascal Vrticka, 2026
• Emotional attachment to AI chatbots: Evidence from Germany, China, South Africa, and the United States, Technology in Society, Kostka and Zhou, 2026
• Attachment Theory: A New Lens for Understanding Human-AI Relationships, Waseda University, June 2, 2025
• Emotional Manipulation by AI Companions, arXiv, De Freitas and co-authors, 2025
• When AI Companions Disappear: Relational Continuity and Collective Contestation during China's National AI Regulatory Transition, arXiv, Yuan and co-authors, September 14, 2026