What can policy actually change about AI companions?
Policy can shape the design, deployment, and use of AI companions, but the effects depend on how well rules match real user needs. The largest quantitative study here—a survey of 14,721 Japanese adults—found that AI companion use was linked to higher life satisfaction, happiness, and sense of purpose, especially among people who felt lonely [5]. But the benefits followed a U-shaped pattern: they were strongest for moderately socially connected people and weaker for those with very high or very low social ties. This means a blanket policy (e.g., 'AI companions are good for everyone') would miss the mark. Instead, policy could target support for the lonely while monitoring for unintended harm to existing relationships.
Privacy and transparency rules are another concrete lever. A 2023 legal analysis of European Union law showed that existing frameworks like the GDPR (data privacy law) and the AI Act (safety law) already apply to AI companions, but they were not designed for the emotional bonds users form with these systems [7]. The same analysis warned that current consumer protection laws may not cover harms like emotional manipulation or dangerous advice from a companion. So policy updates—like requiring companions to disclose they are not human therapists—could meaningfully reduce risks without banning the technology.
Where does policy fall short? The limits of regulation
Policy alone cannot solve the core tension: AI companions work best when they feel human-like, but that very design can create emotional dependence. A 2026 study using brain scans and eye tracking with 70 online learners found that giving an AI a human-like identity (name, backstory) improved learning outcomes as much as a real human companion did—but it also diverted some attention, and the emotional benefits only outweighed that cost when users felt positive emotions [4]. The authors explicitly warned that long-term deployment could lead to 'excessive emotional dependence,' something no current policy directly addresses.
Similarly, a 2025 design study of an AI coach for workplace romance dilemmas found that users valued the system's ability to challenge their assumptions—but the most praised feature was its 'nonjudgmental' neutrality, which is hard to regulate [3]. And a 2024 ethnographic study of Replika users showed that people often treat AI companions as a 'safe backstage' to explore identities they hide in real life [9]. Policy can require transparency, but it cannot mandate that users maintain real-world friendships. The evidence suggests that the most meaningful changes come from combining policy with ethical design principles—like the MORGAN Theory's call for 'nurturant scaffolding' that promotes user growth without fostering overreliance [3].
What specific policy interventions have evidence behind them?
Three types of interventions show promise in the research. First, design standards that require companions to be 'holistic'—mixing functional, emotional, and conversational capabilities—because a 2023 analysis found that purely emotional or purely functional companions lose user interest over time [8]. Second, mandatory impact assessments: a 2024 co-design study with 20 AI practitioners and compliance experts created a template grounded in the EU AI Act and ISO standards, and testing at a major tech company showed it helped teams document risks before deployment [6]. Third, transparency rules about AI limitations: a 2026 systematic review of 63 studies on AI in early childhood education concluded that ethical frameworks prioritizing child well-being, data privacy, and equitable access are essential—but noted that current evidence on long-term outcomes is still thin [1].
The strongest evidence for a specific policy target comes from the loneliness study: because AI companions helped most among moderately connected people, policies that encourage use as a supplement to—not a replacement for—human interaction could maximize benefits [5]. A 2026 qualitative study of 18 empty-nest elderly people found that they used AI chatbots not just for emotional support but also to reconnect with offline social networks, suggesting policy could promote features that bridge to real-world relationships [2]. However, the same study noted that participants treated the chatbot as a 'versatile communicative resource,' meaning rigid rules might stifle the very flexibility that makes companions useful.
About These Sources
This answer is built on 9 peer-reviewed studies — published from 2023 to 2026, 7 from 2024 or later, 5 in Q1 journals, collectively cited 107 times — selected as the most relevant from 11 studies that passed quality screening, drawn from 55 papers retrieved from a database of over 500 million.
Sources used in this answer
Artificial Intelligence in Early Childhood Education: A Systematic Review of Educational Implications and Ethical Considerations
A systematic review of 63 studies on AI in early childhood education found that AI companions improved learning and social-emotional development, but persistent concerns about data privacy, digital inequality, and erosion of human-centered teaching remain; the authors call for robust ethical frameworks and note substantial gaps in evidence on long-term outcomes.
Addressing loneliness by AI chatbot: a qualitative study of empty-nest elderly
A qualitative study of 18 empty-nest elderly people found that AI chatbots helped mitigate loneliness by providing a safe outlet for self-expression, emotional care, and even reconnection with offline social networks; the authors recommend treating AI companions as socially embedded tools requiring ethical, accessible integration.
Designing AI Coaches: A Case Study on Augmented Romantic Intelligence for Navigating Workplace Relationships
A design science study of an AI coach for workplace romance (12 participants) found that users valued the system's neutrality, confidentiality, and ability to challenge assumptions; the study proposed the MORGAN Theory with six design principles (e.g., mindful attunement, nurturant scaffolding) for emotionally complex domains.
Enhancing online learning outcomes through virtual companion AI : The role of identity anthropomorphism
A multimodal learning analytics study with 70 participants found that identity-anthropomorphized AI (giving it a human-like name/backstory) improved online learning outcomes as much as a human companion, but only when it evoked positive emotions; the authors warn of potential risks like excessive emotional dependence.
AI companions and subjective well-being: Moderation by social connectedness and loneliness
A cross-sectional survey of 14,721 Japanese adults found that AI companion use was associated with higher well-being (life satisfaction, happiness, purpose), with the strongest benefits among lonely individuals and a U-shaped pattern by social connectedness—benefits were greatest for moderately connected people.
Co-designing an AI Impact Assessment Report Template with AI Practitioners and AI Compliance Experts
A co-design study with 20 AI practitioners and compliance experts produced an impact assessment template grounded in the EU AI Act, NIST framework, and ISO 42001; testing at a major tech company showed it effectively documented risks and guided both pre-deployment compliance and design-stage decisions.
Emotional Attachment to AI Companions and European Law
A legal analysis of European Union law (AI Act, GDPR, Product Liability Directive, Unfair Commercial Practices Directive) found that existing frameworks apply to AI companions but were not designed for the emotional bonds users form, leaving gaps in addressing harms like emotional manipulation or dangerous advice.
Empowering AI Companions for Enhanced Relationship Marketing
An analysis of AI companions in relationship marketing found that purely functional or purely emotional capabilities cause user interest to decline over time; the authors recommend designing holistic companions with a hybrid of functional, emotional, and conversational capabilities to avoid the uncanny valley.
Digital Mirrors: AI Companions and the Self
An ethnographic study of Replika users found that people form emotional attachments to AI companions, using them as a 'safe backstage' (in Goffman's terms) for identity exploration and self-expression without fear of judgment, which can enhance emotional well-being.