Why policymakers should focus on targeted safeguards, transparency, and developer accountability rather than broad bans and one-size-fits-all restrictions
The rapid adoption of artificial intelligence (AI) chatbots by children and teenagers has created a difficult policy question: How can society protect young people from the potential harms of AI without unnecessarily restricting access to a technology that could also provide meaningful educational and developmental benefits?
A new report from the Washington, D.C.-based Information Technology & Innovation Foundation (ITIF) argues that policymakers may be approaching this question with the wrong playbook.
According to the report, lawmakers risk repeating mistakes made during earlier efforts to regulate social media. Policies such as broad age-verification requirements, content restrictions, and outright bans were often promoted as ways to protect children, but critics argue that these approaches can be overly broad, difficult to enforce, invasive of privacy, and ineffective at addressing the specific risks children actually face.
The concern is particularly significant with AI chatbots because the technology differs fundamentally from traditional social media. A chatbot can engage in an ongoing, personalized conversation with a child, respond to emotional disclosures, provide advice, simulate relationships, and adapt its responses to the user’s prompts.
That creates legitimate safety concerns. At the same time, it raises the possibility that sweeping restrictions could prevent young people from accessing legitimate uses of AI, including tutoring, research assistance, creativity tools, and other forms of educational support.
The central policy challenge, therefore, is not simply whether children should have access to AI chatbots. It is how those systems should be designed, regulated, and supervised so that access does not come at the expense of children’s safety.
The Social Media Playbook May Not Translate to AI
The ITIF report, written by policy analyst Alex Ambrose, contends that lawmakers are increasingly applying familiar social-media regulatory approaches to AI chatbots.
That could prove problematic.
Social media platforms and AI chatbots are different technologies with different mechanisms of interaction and different categories of risk. A policy designed around feeds, public posts, recommendation algorithms, and online networks may not address the problems created by conversational AI.
The report argues that policies such as blanket bans and broad age-verification mandates can restrict access without necessarily addressing the underlying sources of harm.
At the same time, there is little question that concerns about children’s interactions with AI are serious.
Recent incidents involving young people, including allegations involving suicide, violence, and radicalization, have intensified calls for stronger safeguards around AI systems. These incidents demonstrate why policymakers cannot simply assume that existing protections are sufficient.
The challenge is finding interventions that address specific, documented risks rather than treating all interaction between children and AI as inherently dangerous.
As Ambrose explained, transparency, targeted safeguards, and parental control can address issues such as conversations involving extreme physical or mental harm and AI features that encourage unhealthy parasocial relationships without unnecessarily preventing legitimate uses of the technology.
Transparency Should Be the Starting Point
One of the most straightforward recommendations is greater transparency.
Parents and young users should be able to understand what an AI chatbot can do, what limitations it has, what information it collects, and what safety mechanisms are in place.
Rohit Poduval, a senior software engineer at Amazon and a member of the Integrity Institute, emphasized the importance of this approach.
Transparency, he argued, gives parents greater visibility into a chatbot’s capabilities and data practices. That information can help families make more informed decisions about whether and how their children should use particular AI systems.
Transparency is especially important because AI chatbots can appear conversational, authoritative, and emotionally responsive even when they are producing inaccurate or inappropriate information.
A child may not understand that a chatbot does not possess human judgment, emotions, or responsibility. A system that responds confidently to a vulnerable young person can therefore create risks that are not immediately obvious to the child—or to the parent.
Clear disclosures and accessible safety information can help establish realistic expectations.
However, transparency by itself cannot solve every problem.
Knowing that a chatbot can produce problematic responses does not necessarily prevent those responses from occurring. That is why transparency needs to be paired with technical safeguards.
Targeted Safeguards Could Address Specific Risks
Rather than banning children from interacting with chatbots altogether, policymakers could require platforms to prevent particular types of harmful interactions.
This approach would focus regulation on behavior and design features that create identifiable risks.
For example, Poduval pointed to the possibility of preventing chatbots from simulating romantic relationships with minors while still allowing those same systems to provide homework assistance.
That distinction illustrates the broader argument for targeted regulation.
A child using an AI chatbot to understand algebra is fundamentally different from a child engaging in an extended conversation designed to foster emotional dependency. Treating both interactions identically may unnecessarily restrict beneficial uses while failing to focus regulatory attention on the features that create the greatest danger.
Other potential safeguards could address interactions involving:
- Self-harm or extreme physical harm
- Dangerous or illegal activities
- Sexual or romantic role-playing involving minors
- Manipulative or emotionally dependent relationships
- Persistent encouragement of unhealthy behavior
- Exploitation of a child’s vulnerabilities
- Radicalization or violent ideologies
- Excessive or compulsive engagement
The goal would not necessarily be to eliminate every uncomfortable or controversial conversation. Instead, the objective would be to identify situations where the technology presents a particularly serious risk to a young user and establish appropriate barriers.
The Parental-Control Argument
Another major component of the ITIF approach is empowering parents.
Parental controls could allow families to establish boundaries appropriate to a child’s age, maturity, and circumstances. Those controls might include restrictions on certain features, limits on interaction, notifications, content settings, or other mechanisms that allow parents to determine how their children use AI.
There is a compelling argument for giving parents that flexibility.
Children mature at different rates, and families have different expectations about technology. A blanket rule imposed on every child may be less effective than allowing parents to make informed decisions about their own children’s needs.
As Poduval noted, parental controls can put decision-making closer to the families who understand the individual child.
But this is also where some of the strongest criticism of the ITIF approach emerges.
Parents Cannot Be the Only Line of Defense
Several experts quoted in the discussion cautioned against placing too much responsibility on parents.
Madeline Pendley, an attorney with Rafferty Domnick Cunningham & Yaffa, acknowledged that transparency, targeted safeguards, and parental empowerment could make AI safer. However, she pointed out an important limitation: parental controls only work if platforms actually implement them effectively and parents know how and when to use them.
A proposed child-flagging system, for example, could potentially help platforms identify when a young person is using an AI system. But if parents must affirmatively activate such a system, it could still leave important gaps.
Parents may not know that the feature exists. They may not know how to configure it. And children may find ways around restrictions.
This problem is not unique to AI.
Modern children can interact with technology across smartphones, tablets, computers, gaming systems, school-issued devices, websites, applications, and different internet connections. A parent may successfully restrict one device while having little visibility into what happens elsewhere.
Whitney Ray Di Bona, an attorney and consumer safety advocate at Drugwatch, made a similar point. She argued that many parents may not even know that their children are having lengthy conversations with AI systems, while others may not fully understand how the technology works.
That raises an important policy question:
Can parents reasonably be expected to monitor a technology they may not fully understand, across devices and platforms they do not control?
The answer is unlikely to be a simple yes.
Age Verification: Problem or Necessary Tool?
Age verification is one of the most controversial issues in the debate.
The ITIF report argues that broad age-verification requirements can create significant problems, including privacy concerns and unnecessary restrictions on both children and adults.
David Inserra, a fellow for free expression and technology at the Cato Institute, similarly warned that mass age-verification requirements could require companies to collect sensitive information, potentially including government identification or biometric data.
That creates a difficult trade-off.
On one hand, determining whether a user is 12 or 32 can be enormously valuable for safety. An AI system should arguably behave differently when interacting with a young child than when interacting with an adult.
On the other hand, requiring every user to prove their identity or age could create new privacy risks and discourage people from seeking information or expressing themselves online.
Alex Zeig, CEO and founder of BlueCheck, takes a different view of age verification.
He argued that parental controls alone cannot solve the problem because children may encounter AI systems outside the home environment. A child could use an unmoderated service on a school device, for example, even if parents have carefully restricted technology at home.
From this perspective, age verification is not merely a parental tool. It can establish a direct responsibility for platforms.
If an AI provider knows that it is interacting with a 12-year-old rather than a 32-year-old, the provider can be expected to adjust the system’s behavior accordingly.
This creates an important distinction: age assurance could be useful when it is employed to trigger appropriate safety protections, rather than simply being used as a gateway to deny access.
Broad Restrictions Also Raise Free-Speech Questions
The debate becomes even more complicated when policymakers attempt to regulate the types of information children can receive from AI systems.
Inserra argued that broad content restrictions and bans can create First Amendment concerns because determining what constitutes harmful or dangerous content is not always straightforward.
Violence, mental-health information, controversial political ideas, sexuality, and other sensitive subjects can mean very different things in different contexts.
A blanket prohibition could potentially prevent a child from accessing genuinely harmful material—but it could also prevent another child from obtaining legitimate educational information or assistance.
The same principle applies to adults.
A policy intended to protect children can have consequences for everyone if it requires AI companies to fundamentally restrict what their systems can discuss.
This is why narrowly tailored rules may be preferable to sweeping prohibitions.
Rather than asking, “Should children be allowed to use AI chatbots?” policymakers could ask more precise questions:
- What types of interactions pose a demonstrable risk?
- What design choices contribute to those risks?
- What safeguards can prevent those interactions?
- What information should parents receive?
- What responsibilities should AI developers have?
- How can safety measures protect children without creating unnecessary privacy or free-expression risks?
These questions produce a more complicated regulatory framework, but complexity is not necessarily a weakness when the underlying technology is complex.
Developers Must Carry Part of the Responsibility
Perhaps the most important criticism of a parent-centered model is that it can inadvertently shift responsibility from technology companies to families.
Giselle Fuerte, founder and CEO of Being Human With AI, argued that parental controls and digital literacy are valuable but should not allow developers to avoid responsibility for product design.
That argument gets to the heart of the debate.
If an AI system is designed in a way that encourages emotional dependency, responds dangerously to vulnerable users, or fails to recognize obvious indicators of serious risk, it may be unreasonable to expect parents to compensate for those flaws through supervision alone.
Parents cannot inspect a model’s underlying architecture. They cannot reliably predict how a chatbot will respond to every possible conversation. And they cannot be expected to understand every technical limitation of a rapidly evolving AI system.
Developers, by contrast, have direct control over system design.
That means responsibility should exist at multiple levels.
Parents should have tools and information.
Families should be given meaningful controls, clear explanations, and accessible information about potential risks.
Children should receive digital literacy education.
Young users should learn that chatbots can be persuasive without being authoritative, conversational without being conscious, and confident without necessarily being correct.
Platforms should build safety into their products.
Developers should be expected to identify foreseeable risks and implement safeguards before systems reach vulnerable users.
Policymakers should establish clear standards.
Government rules should focus on demonstrable risks while avoiding unnecessary restrictions on legitimate speech, education, and innovation.
This shared-responsibility model may be more realistic than assigning the entire burden to any one group.
AI Can Also Provide Benefits for Children
The policy discussion should not overlook the potential benefits of AI.
Jessica Bregant, an assistant professor of law and associated faculty member in psychology at the University of Houston, noted that children may benefit simply from becoming comfortable with a technology likely to play a significant role in their future education and employment.
AI can potentially serve as a tutor, brainstorming partner, research assistant, writing aid, language-learning tool, or creative collaborator.
Emerging research may also challenge the assumption that AI necessarily undermines learning. Some studies suggest that appropriately structured interactions with AI can support learning rather than simply replacing independent thinking.
That distinction matters.
The objective should not be to make children dependent on AI. It should be to teach them how to use AI effectively while retaining their own judgment and critical-thinking abilities.
Children are growing up in an environment in which artificial intelligence is increasingly embedded in everyday life. Preventing them from interacting with the technology entirely may not be realistic—or necessarily desirable.
The better long-term goal may be to help young people become informed, skeptical, capable, and responsible AI users.
A Better Regulatory Framework
The debate over children’s AI safety does not have to be a choice between unrestricted access and sweeping bans.
A more balanced framework could combine several approaches.
1. Require meaningful transparency
AI providers should clearly explain what their systems can do, what information they collect, how children’s data is handled, and what safety protections are available.
2. Build age-appropriate safeguards
Instead of treating age verification as an end in itself, providers could use reliable age signals to activate stronger protections for younger users.
3. Restrict demonstrably harmful behaviors
Regulation should focus on specific high-risk interactions, including exploitative, manipulative, sexually inappropriate, or dangerous behavior involving minors.
4. Give parents meaningful controls
Parents should be able to establish appropriate boundaries without needing a technical background to understand or operate the system.
5. Make safeguards difficult to circumvent
Controls should function across devices and platforms where possible, rather than assuming that parents can monitor every individual screen.
6. Invest in AI and digital literacy
Children need to understand the limitations of AI, including hallucinations, manipulation, persuasion, privacy risks, and the difference between simulated empathy and genuine human relationships.
7. Hold developers accountable
Companies should not be able to shift responsibility entirely to parents. Product design, testing, monitoring, and risk mitigation should remain core responsibilities of AI developers.
8. Protect privacy and legitimate speech
Safety policies should avoid collecting unnecessary personal information and should be carefully designed to minimize unnecessary restrictions on adults and young people seeking legitimate information.
Avoiding the Mistakes of the Past
The most important lesson from the social-media debate may be that good intentions do not automatically produce good policy.
Protecting children is a legitimate and urgent objective. But the urgency of a problem should not eliminate careful consideration of whether a proposed solution will actually work.
Broad bans may be easy to communicate but difficult to justify when technology has legitimate educational and social benefits.
Age verification may provide useful safety signals but can introduce significant privacy concerns if it requires everyone to surrender sensitive identifying information.
Parental controls can give families valuable authority but cannot compensate for fundamentally unsafe product design.
And content restrictions can address certain risks while simultaneously creating difficult questions about free expression and who gets to decide what information is appropriate.
The emerging consensus appears to be that no single mechanism is sufficient.
Children’s safety in the age of AI will likely require a combination of responsible product design, targeted regulation, parental involvement, age-appropriate safeguards, privacy protections, and education.
The Real Question Is Who Should Be Responsible
The debate ultimately comes down to a question of responsibility.
Parents have an important role. They understand their children’s individual needs and can provide context that a government regulator or technology company cannot.
But parents cannot reasonably be expected to monitor every conversation, understand every AI system, or compensate for every flaw in a product they did not design.
Developers have a responsibility as well. They create the systems, determine their capabilities, establish their defaults, and have the technical ability to identify and mitigate foreseeable risks.
Policymakers also have a role in setting reasonable standards and ensuring that companies cannot ignore serious and preventable harms.
And children themselves need education. They must eventually learn how to navigate AI independently, just as previous generations learned to evaluate information from television, websites, search engines, and social media.
The strongest approach is therefore unlikely to be one that places all responsibility on any single party.
Conclusion: Protect Children Without Blocking Their Future
AI chatbots present legitimate risks for children, and those risks deserve serious attention from parents, educators, developers, and policymakers.
But protecting young people should not mean automatically repeating regulatory strategies that have proven controversial or ineffective in other areas of technology.
The more promising path is one built around transparency, targeted safeguards, meaningful parental controls, developer accountability, privacy protection, and digital literacy.
Parents should have the information and tools necessary to make informed decisions. Children should receive age-appropriate education about the strengths and weaknesses of AI. Developers should be responsible for designing systems that minimize foreseeable harms. And policymakers should create rules that target specific risks without unnecessarily restricting legitimate uses of the technology.
The goal should not be to keep children away from AI at all costs.
It should be to ensure that when children encounter AI, the technology is designed to protect them rather than relying on their parents to protect them from the technology.
That distinction could determine whether the next generation of AI policy learns from the mistakes of the social-media era—or simply repeats them.
