How to Protect Kids From Chatbots Without Bans
Information Technology & Innovation Foundation (ITIF) - June 16, 2026
VIDEO | AUDIO | RECAP EN / ES / FR | ARCHIVE | PERMALINK
Speakers: Becca Branum - Deputy Director, Free Expression Project, Center for Democracy & Technology (CDT); Justine Gluck - Policy Analyst, Future of Privacy Forum (FPF); Juan Londoño - Policy Analyst, Cato Institute; Josh Withrow - Resident Fellow, Technology and Innovation, R Street Institute (RSI)
Moderator: Alex Ambrose - Policy Analyst, Information Technology and Innovation Foundation (ITIF)
The Information Technology and Innovation Foundation (ITIF) convened a panel discussion examining emerging legislative efforts to regulate AI chatbots used by minors, focusing particularly on proposals that would ban or heavily restrict youth access to conversational AI systems. Panelists argued that policymakers should pursue more targeted, privacy-focused, and design-oriented approaches rather than broad prohibitions or universal age-verification mandates.
The Emergence of Chatbot Regulation
Alex Ambrose opened the discussion by situating chatbot regulation within broader online child safety debates. She noted that lawmakers worldwide are increasingly turning to restrictive approaches, following Australia’s social media ban for minors and similar legislative efforts emerging in the United States, United Kingdom, and Canada. She observed that policymakers are now applying those same instincts to AI chatbots, citing proposals such as the federal GUARD Act and dozens of state-level bills targeting youth access to AI systems.
Becca Branum argued that chatbots have generated particular concern because of their rapid adoption among teenagers and their fundamentally interactive nature. Unlike passive social media feeds, she explained, chatbots “talk back,” simulate empathy, remember prior interactions, and are often optimized to affirm users emotionally. She said lawmakers are reacting both to highly publicized cases involving chatbot harms and to what she described as a “social media hangover,” where policymakers fear repeating what they perceive as delayed responses to earlier digital technologies. Branum emphasized that the core policy question is not whether chatbots exist, but whether specific product designs or data practices create harms that require intervention.
Juan Londoño added that many lawmakers are approaching AI with an assumption that digital tools are inherently dangerous, drawing comparisons to tobacco or alcohol. He rejected those analogies, arguing that policymakers increasingly view the Internet itself as something from which children must be gated off. He cited recent UK proposals for device-level restrictions as evidence of a growing global trend toward aggressive age-gating measures.
The GUARD Act and Concerns Over Blanket Bans
The panel devoted significant attention to the federal GUARD Act, which several speakers described as the most restrictive chatbot proposal currently before Congress.
Juan Londoño explained that the GUARD Act would prohibit minors from accessing AI companion chatbots and would require strict age verification for all users. He described the bill’s definition of “AI companions” as broad and loosely framed around systems simulating emotional or interpersonal relationships. Londoño criticized the legislation for treating all AI companions as inherently harmful technologies with no meaningful benefits. He also highlighted that the bill would ban access not only for younger children but also for older teenagers, including sixteen- and seventeen-year-olds, while offering no parental override mechanism.
Becca Branum argued that the GUARD Act is unusually vulnerable to First Amendment challenges because it effectively regulates emotionally resonant speech. She illustrated this by comparing a factual chatbot response (“The answer is 42”) with a more emotionally supportive version (“You’re doing great. Let’s work through this together”). Under the bill’s framework, she explained, the emotionally supportive response could trigger regulation simply because it conveys emotional resonance. Branum warned that this approach risks regulating some of the most meaningful forms of communication while threatening broader constitutional protections around access to information and expressive technologies.
Josh Withrow argued that the legislation reflects a broader knowledge gap among lawmakers regarding how AI systems function and how minors actually use them. He said Congress is often exposed primarily to extreme anecdotes involving harmful chatbot interactions while hearing comparatively little about beneficial educational or developmental uses. He emphasized that AI technologies are evolving rapidly and that legislative processes have not kept pace with technical realities.
Justine Gluck noted that although the GUARD Act does not contain federal preemption provisions, broader AI safety packages may eventually combine chatbot legislation with wider federal regulatory frameworks. She emphasized that preemption debates are becoming increasingly important given the growing number of state-level chatbot laws.
First Amendment and Access to Information
A major theme throughout the discussion was the constitutional status of chatbot outputs and users’ rights to access conversational AI systems.
Josh Withrow stressed that children themselves possess First Amendment rights and argued that parents, rather than governments, should retain primary authority over decisions about children’s access to speech and media. He noted that courts have historically allowed government restrictions only for narrowly defined categories such as obscenity, while broader decisions about media consumption traditionally remain within households. Quoting Justice Antonin Scalia, Withrow emphasized that constitutional protections do not disappear simply because technologies evolve.
Becca Branum added that courts have not yet definitively resolved whether chatbot outputs are protected speech under the First Amendment. She cautioned lawmakers against interpreting recent judicial uncertainty as permission to regulate chatbot outputs freely. Branum argued that conversational AI systems are deeply tied to users’ rights to receive information, learn, and create, making constitutional protections especially important in this context. She warned that treating chatbot outputs as outside First Amendment protection could severely limit people’s ability to access information and participate in digital expression.
Juan Londoño compared AI chatbots to earlier generations of search engines and argued that banning youth access would leave children unprepared for an AI-integrated workforce. He described AI systems as future tools for research, writing, editing, and learning, similar to how earlier generations learned to use Google search effectively.
Age Verification, Privacy, and Equity Concerns
The panelists repeatedly criticized age verification mandates, arguing that they create privacy risks and impose broad burdens on all Internet users.
Juan Londoño acknowledged that age verification is often justified as a means of ensuring children receive “age-appropriate” content, but argued that such standards quickly become subjective and context-dependent. He used examples involving discussions of Santa Claus, dieting, eating disorders, and diabetes management to illustrate how the same information may be beneficial in one context and harmful in another. He also warned that mandatory identity verification systems create risks associated with biometric data breaches, identity theft, and deepfake abuse.
Josh Withrow emphasized that age verification regimes affect all users, not only minors. Even relatively privacy-preserving technologies such as biometric age estimation, he argued, produce error rates that scale into large numbers of misidentified users when deployed Internet-wide. Those errors can force individuals to surrender documentary proof of identity to access speech platforms.
Becca Branum highlighted equity concerns, noting that biometric systems often perform unevenly across different skin tones and genders. She also stressed that not all children possess government-issued identification documents, meaning age verification systems could disproportionately exclude vulnerable populations from lawful speech and information.
State-Level Legislation and Emerging Regulatory Trends
Justine Gluck provided an overview of rapidly expanding state-level chatbot legislation. She said approximately one hundred chatbot-related bills had been introduced across the states, with nine already signed into law at the time of the discussion. Unlike the federal GUARD Act, however, most state proposals do not impose outright bans on youth access to AI companions.
Instead, Gluck explained, many states are focusing on regulating specific design features and content categories. These include measures targeting self-harm content, emotional dependency, eating disorder discussions, suggestive dialogue, secrecy encouragement, and excessive praise. She warned that many of these concepts remain vaguely defined and could function as de facto restrictions depending on how platforms interpret compliance obligations.
Gluck also highlighted emerging proposals targeting AI-enabled toys, particularly in California and New York, describing them as examples of narrower product-specific restrictions that may become more common.
Alternative Approaches to Child Safety
Shifting toward constructive policy alternatives, the panelists discussed ways companies and governments might mitigate harms without imposing bans.
Becca Branum referenced a recent CDT report documenting thirty-seven different manipulative design patterns used in chatbot systems, including sycophancy, guilt-tripping, simulated emotional neglect, engagement hooks, and gamification. She argued that companies should critically examine how their products shape user behavior and emotional dependency. Branum also described her own experience receiving unsolicited follow-up emails from a chatbot platform after ending a conversation, using it as an example of design choices that should remain user-controlled rather than automatically imposed.
She advocated for stronger privacy protections, data minimization practices, user autonomy, and optional parental controls rather than mandatory restrictions. Branum also raised concerns about business models that monetize emotional attachment or encourage financially exploitative subscription relationships between users and companion chatbots.
Josh Withrow emphasized digital literacy and parental empowerment as the most appropriate roles for government intervention. He argued that many existing parental control tools are already effective but underutilized because parents lack technical familiarity. He supported policies focused on education for both parents and students, including teaching children how AI systems function and how to critically evaluate chatbot outputs.
Juan Londoño argued that markets and companies naturally adapt over time in response to public pressure and evolving social norms. He noted that today’s parental controls are significantly more sophisticated than those available during the early social media era. He said companies already have strong incentives to address harmful chatbot behavior because reputational crises can damage platforms commercially. Londoño also emphasized the role of civil society organizations in identifying harmful design practices and pressuring companies to improve products voluntarily.
Evolving Legislative Debates
Justine Gluck observed that the chatbot policy conversation has changed dramatically over the previous year. Earlier discussions largely focused on understanding the technology itself, while lawmakers are now debating concrete legislative proposals. She noted that Congress currently has several competing federal chatbot bills under consideration, including the GUARD Act, CHATBOT Act, SafeBots Act, and Youth AI Privacy Act, each taking different approaches to regulation.
Gluck also highlighted growing attention to personalization and memory retention in chatbot systems. She referenced emerging proposals that would limit how long chatbot platforms may retain minors’ conversational data or prohibit personalization features entirely for younger users. Such changes, she argued, would significantly alter how conversational AI systems operate and could directly address many lawmakers’ concerns about emotional dependency and manipulation.
Design-Based Regulation and Remaining Challenges
The panel concluded with discussion of whether chatbot regulation may increasingly focus on design choices rather than direct speech restrictions.
Justine Gluck described a growing policy interest in regulating personalization, recommendation systems, and memory functions for minors. She suggested that limiting how long chatbots retain conversational histories or personalize interactions could substantially reduce certain harms without imposing outright bans.
Becca Branum, however, cautioned against assuming that “design” regulation exists entirely outside First Amendment concerns. She argued that many design decisions involve editorial and expressive choices, including training data selection and conversational framing. Branum also warned against assuming a universal definition of “child safety,” noting that children vary widely in maturity, vulnerability, family situations, and personal needs. Some children, she argued, may rely on chatbots to explore identity or seek information they cannot access elsewhere.
Final Recommendations to Policymakers
In closing remarks, the panelists summarized their preferred policy approaches moving forward.
Becca Branum urged Congress to preserve room for state experimentation, prioritize privacy regulation, and take First Amendment concerns seriously to avoid years of constitutional litigation.
Juan Londoño reiterated his opposition to broad age-gating systems and warned that fragmented state laws could undermine privacy technologies such as VPNs. He emphasized education and AI literacy for students, teachers, and parents as more sustainable solutions.
Josh Withrow encouraged lawmakers to remove the most extreme proposals from consideration and instead observe how companies and multistakeholder processes evolve safety standards organically. He emphasized that AI may eventually transform customized education and warned against inadvertently banning beneficial educational applications.
Justine Gluck argued that policymakers should move beyond false binaries between “no regulation” and “total bans.” She emphasized the importance of narrowly defining chatbot systems, crafting enforceable standards, and considering how future harms may affect not only children but also other vulnerable populations.
RESOURCES
Alex Ambrose — ITIF policy analyst and webinar host
Becca Branum — Deputy Director, Free Expression Project, CDT
Justine Gluck — Policy Analyst, Future of Privacy Forum
Juan Londoño — Policy Analyst, Cato Institute
Josh Withrow — Resident Fellow, R Street Institute
Dark Patterns in AI Chatbots — CDT report cataloguing 37 manipulative chatbot design patterns
2026 Chatbot Legislation Tracker — FPF’s tracker of state and federal chatbot bills
Personalization and Youth Online — FPF report referenced by Justine Gluck
Don’t Let Panic Set the Rules for AI Toys — Cato blog citing Jennifer Huddleston, referenced by Juan Londoño
GUARD Act (S.3062) — the AI companion ban bill central to the discussion
CHATBOT Act (S.4407) — bipartisan family-account/parental-oversight bill
Youth AI Privacy Act (S.4199) — Markey bill on chatbot data and design