<p>The rapid proliferation of artificial intelligence (AI) companions and social chatbots has reshaped human emotional landscapes, raising urgent questions about intimacy, ethics, and mental health in human-AI relationships. This AI-assisted systematic review synthesises 37 studies (2021–2025) to examine how AI mediates intimacy, balancing its potential to alleviate loneliness with risks of dependency and privacy violations. Guided by Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines, we identify four core themes: (1) emotional and social mechanisms, such as anthropomorphism and identity exploration; (2) dual psychological impacts, including reduced isolation versus over-reliance; (3) ethical dilemmas like data exploitation and algorithmic manipulation; and (4) cultural variation in adoption, such as therapeutic acceptance in East Asia compared to authenticity debates in Western contexts. Key findings reveal that AI companions, such as Replika, simulate trust and reciprocity through personalised interactions, yet users often grapple with authenticity dilemmas and emotional dissonance. Anchored in attachment theory and socio-affective frameworks, the analysis highlights emerging risks, such as AI-induced role-reversal and pseudo-intimacy, while underscoring design and policy imperatives. We propose actionable safeguards: transparency in AI’s emotional limitations, culturally nuanced regulations, and user-controlled ‘emotional break’ features to mitigate harm. By bridging interdisciplinary perspectives, this review advances a global roadmap for ethical AI design that prioritises human dignity alongside technological innovation.</p> Graphical Abstract <p></p>