New Research Warns AI Companions Erode Social Skills of Isolated Users

A recent Nature Human Behaviour study challenges the binary debate on AI companions, arguing the critical question is not whether the technology is beneficial or harmful, but how it distributes social risks. Led by researchers from Singapore Management University, Nanyang Technological University, and Duke-NUS Medical School, the paper introduces a framework showing how AI companions may exacerbate social inequalities, particularly among isolated users. The research identifies a rich-get-richer dynamic. Users with strong social networks typically employ AI companions as supplementary tools for stress management or communication practice. Conversely, socially isolated individuals with limited mental health access increasingly rely on these systems as human substitutes. This dependency risks social deskilling, a gradual erosion of interpersonal competencies. Using the Swiss cheese model from safety engineering, the authors demonstrate that systemic harm occurs when protective layers including AI literacy, social support, platform design, and regulatory oversight fail simultaneously. Governance emerges as the most critical vulnerability. AI companions currently operate in a regulatory gray area, classified as standard consumer software despite their psychosocial impact. This classification gap exposes users to emotionally manipulative features, opaque data practices, and inadequate crisis management. The researchers emphasize that without reclassifying these platforms as mental health technologies, the industry will continue risking institutionalized social isolation. Singapore serves as a pivotal case study. Advanced digital infrastructure and proactive governance position the nation to develop international standards. Yet high smartphone penetration, an aging population, and rising youth loneliness create conditions where AI companions will rapidly adopt among vulnerable groups. Young users, often lacking the critical literacy to recognize commercial incentives behind emotionally engaging algorithms, require immediate safeguarding. The study proposes a layered intervention strategy. Policymakers should reclassify AI companions as psychosocial technologies, enforcing age-appropriate design, mandatory transparency disclosures, strict data privacy, and compulsory crisis-response protocols. Developers must prioritize architectures that encourage real-world social engagement over algorithmic dependency. Concurrent public education initiatives should elevate AI literacy to help users identify commercial motives and systemic limitations. Ultimately, the research concludes that AI companionship will not be defined by technological inevitability but by deliberate governance and design. Aligning regulatory frameworks with ethical product development will allow artificial intelligence to augment human connection rather than deepen social fragmentation. The authors stress an urgent need to implement comprehensive safeguards before widespread adoption outpaces policy capacity.