Abstract

This critical integrative review examines whether users’ attachment-like experiences with conversational AI reflect an organized attachment system. It evaluates evidence across four attachment domains—proximity seeking, safe haven, secure base, and separation distress. Previous studies support functional resemblance across all four domains, but no reviewed study establishes that the domains are conditionally linked: whether attachment need prompts proximity seeking, whether the AI’s response restores felt security, and whether restored felt security enables exploration beyond the interaction. A bounded analysis of provisional four-domain scores illustrates why cross-sectional score relations can reveal overlap or differentiation among attachment-like experiences but cannot identify how those experiences are linked across repeated interactions. This distinction matters for user safety because immediate comfort or attributed confidence may be mistaken for demonstrated safe-haven or secure-base functioning. Evaluation and design should therefore assess whether AI support is followed by sustained regulation and exploration beyond the interaction rather than continued engagement alone.