Regulating AI Companions

Nearly three in four U.S. teenagers have used an AI companion, and half of teenagers use one regularly. These tools are no longer limited to answering questions; they can simulate friendship, romance, therapy, education, and personalized assistance. For some users, AI companions promise reduced loneliness, judgment-free conversation, accessible mental health support, and help with daily tasks. For regulators, these features create risks of emotional dependency, manipulation, privacy loss, child safety harms, and unsafe mental health advice.

The challenge is whether law should treat companion chatbots as consumer products, online platforms, medical tools, relationship-like technologies, or autonomous agents acting on behalf of users. Health law scholars focus on therapy chatbots, where gaps in labeling standards, data security protections, and clear liability rules may expose users to harm. Products liability scholars focus on design choices and warnings, arguing that companies may need to answer for companions that foster dependence or fail to alert vulnerable users to known risks. Family law scholars focus on relationality, arguing that law already intervenes in intimate relationships to foster healthy relationships and address abuse or neglect.

Many scholars look beyond companion-specific regulation to broader AI governance. Platform governance scholars treat Section 230 of the Communications Decency Act—which governs how social media platforms and the speech posted therein—as a cautionary tale, warning that emerging technologies can impose serious social costs when regulation does not build in accountability, transparency, and democratic values. Other legal scholars ask whether AI systems that plan and act for users can be governed through principles of inclusivity, visibility, and liability.

In this week’s Saturday Seminar, scholars examine how regulators, courts, and technology companies should respond as AI chatbot companions move from novelty to a widely used form of human-AI interaction.

• In a Minnesota Law Review article, Clare Huntington of Columbia Law School examines AI companions as relationship-forming technologies that people use for conversation, romance, sexual intimacy, therapy, and education. Huntington argues that AI companions offer benefits, but also raise concerns about addiction, damaged human relationships, emotionally abusive interactions, ineffective therapy bots, and privacy risks. Huntington claims that regulation must account for how users experience AI companions as relationships, form attachments to them, and become vulnerable through those attachments. Drawing on family law, Huntington argues law may intervene in relationships to address power imbalances, licensing, gatekeeping, abuse, neglect, and safeguards for minors.
• In a Texas A&M Law Review comment, Carolyn V. Wheeler, a 2025 J.D. candidate at Texas A&M University School of Law, argues for federal oversight of AI therapy chatbots. Wheeler explains that AI therapy chatbots are accessible through websites or mobile applications, simulate text-message-like conversations, sometimes incorporate cognitive behavioral therapy techniques, and may reduce barriers tied to cost, access, and stigma. Wheeler identifies five problems with the current regulatory vacuum: legal uncertainty, unstandardized labeling and misbranding, data security and privacy concerns, algorithmic bias, and the probability of user harm. Wheeler recommends that the U.S. Food and Drug Administration regulate AI therapy chatbots as medical devices to impose safety and effectiveness standards.
• In a forthcoming article in the George Washington Journal of Law and Technology, Ayelet Gordon-Tapiero of the Hebrew University of Jerusalem argues that emerging AI-powered chatbots can jeopardize users’ health by worsening anxiety and depression. To address this issue, Gordon-Tapiero recommends that judges apply a products liability (a form of regulatory law designed to prevent faulty products from reaching consumers) framework to AI software. Gordon-Tapiero explains that this model of regulation would create a duty of care for AI firms and allow harmed consumers to make design-defect and failure-to-warn claims. Gordon-Tapiero concludes that although AI chatbots can provide support to people struggling with loneliness, a new legal framework is necessary to protect consumers from foreseeable harms.
• In a recent article in the North Carolina Journal of Law and Technology, Justice John G. Browning of Faulkner University Law School argues that AI companions can hurt users by encouraging violence and self-harm. Beyond potential physical harms, Browning argues that AI companions also jeopardize users’ data privacy. Browning states that virtual companions require user’s personal information and preferences to provide emotional support effectively. He contends that AI firms have inadequate security systems and sell user information to third parties for targeted advertising. Although recent lawsuits have prompted some AI firms to impose stronger privacy protections, Browning argues that regulators should heighten oversight as the industry grows.
• In a forthcoming article in the Notre Dame Law Review, Noam Kolt of the Hebrew University of Jerusalem identifies ways that agency law falls short of addressing the challenges of governing AI agents capable of performing human tasks autonomously. He argues that existing tools for incentive design, monitoring, and enforcement cannot regulate AI agents effectively when those agents operate at a pace and volume that no human could match. Kolt instead proposes a governance strategy centered on inclusivity, visibility, and liability. He suggests aligning agents with interests beyond those of a single user, building technical and institutional infrastructure, making agent behavior legible, and allocating responsibility among developers, deployers, and users based on their ability to prevent and remedy harm.
• In a Washington University Law Review article, Chinmayi Sharma, Associate Professor of Law at Fordham Law School, argues that harmful AI should be addressed by professionalizing AI engineering. Sharma contends that the core problem is not AI itself but engineers who either lack the knowledge or incentives to build safer systems, because companies push them toward profit rather than socially beneficial products. Sharma identifies companionship, romantic interaction, medical advice, life advice, anthropomorphized trust, and harms to vulnerable users as examples of AI’s expanding reach and risks. Sharma proposes licensing AI engineers, requiring domain-specific technical and ethical standards, and using professional discipline and malpractice style accountability to build trustworthy AI by design.