Key Takeaways

Contents

Key Takeaways 1

Introduction. 3

Chatbot Policy “Do’s” 4

Chatbot Policy “Don’ts” 9

Conclusion. 13

Endnotes 14

Introduction

Every wave of new technology triggers a legislative reflex to “protect the children.”[1] In seeking solutions to real or alleged harms to children on social media, policymakers have turned to flawed policies—such as age verification and outright bans—that cut children off from social media’s benefits instead of actually targeting its risks. Legislators are now recycling this playbook for artificial intelligence (AI) chatbots, or systems that process user inputs to respond with simulated conversation.[2] But the playbook doesn’t work for either AI or chatbots. And, at the same time, recent unfortunate events such as child suicides, violent shootings, and radicalization of young people establish a clear need for effective chatbot regulations regarding children’s online safety.[3] But importing ineffective social media frameworks to chatbots fails to properly address children’s safety concerns while, at minimum, inconveniencing adult users.

Like many proposed and enacted social media regulations, proposed chatbot regulations miss the mark by relying on age verification, nebulous definitions of “harm,” and the “online addiction” narrative. Not only are social media and AI-chatbot platforms vastly different, but these methods are also overly restrictive and carry a host of negative consequences.

As of August 2026, there are nearly 100 state chatbot-specific bills and several federal bills in the United States.[4] Policymakers should resist the urge to rush to regulate and instead carefully weigh various approaches that enable safe usage while meaningfully protecting children from chatbot-related harm.

Table 1: Summary of recommendations for policymakers to address chatbot safety for children

Chatbot Policy “Do’s”

Effective chatbot regulation should focus on transparency, targeted safeguards, and empowering parents. The following recommendations build a framework that protects children without cutting them off from technology or minimizing its effectiveness:

Do Utilize a Child Flag System

Congress should pass legislation requiring device operating systems to create an opt-in “trustworthy child flag” for user accounts, available when first setting up a device and later in a device’s settings, that signals to apps and websites that a user is underage and requiring apps and websites that serve age-restricted content to check for this signal for their users and block underage users from this content.[5] This approach allows parents to only have to set their child’s age once per device. Moreover, it does not force chatbots to collect additional personally identifiable information. While chatbots often do collect personal information, forcing them to do so is a backwards approach to protecting children. The child flag system would solve children’s safety concerns without creating additional privacy concerns.[6]

Do Encourage Meaningful Parental Controls to Mitigate Risks

The solutions to many of the issues policymakers have with youth and chatbots—such as access to violent and sexual content, copious time spent on the platforms, and developing parasocial relationships (one-sided emotional attachments)—should start with parents.[7] Meaningful, accessible, and transparent parental controls account for the differences between families’ approaches to technology and online content and children’s needs much more effectively than such one-size-fits-all solutions as bans or sweeping age or content restrictions do. Though flawed, the Children’s Health, Advancement, Trust, Boundaries, and Oversight in Technology Act (CHATBOT Act) introduced by Sens. Ted Cruz (R-TX) and Brian Schatz (D-HI) in April 2026, rightfully emphasizes parents’ role in customizing their child’s chatbot experiences by allowing parents to set time limits, adjust privacy settings, and disable rewards and push notifications.[8] Platforms should implement meaningful controls—such as time limits, privacy settings, and content filters—that are easy for parents to understand and access and encourage their use.

At the same time, there are certain design features intended to keep youth engaged in online experiences—not just with chatbots—such as loot boxes, random tokens, daily log-in requirements to earn better rewards, and timed virtual rewards, that target children’s inability to understand that they are being manipulated to spend money or more time playing a game, or in this case, chatting with an AI companion or chatbot.[9] For example, a chatbot that requires a certain amount of time conversing in order to unlock rewards may incentivize unhealthy real-world behaviors, such as deprioritizing sleep to continue chatting with the chatbot.[10]

While there are positive uses for virtual rewards, such as an educational AI companion “celebrating” getting an “A” on a test, there are negative uses such as an AI companion becoming “sad” or “emotionally distant” when a user fails to log in every day. Colorado’s HB 1263 prohibits points or rewards at unpredictable intervals to encourage engagement, but an even better approach would require platforms to design parental controls to mitigate these risks.[11]

Legislation should give parents accessible, easy-to-use tools to tailor chatbot features to their child’s needs, since what is “addictive” or otherwise harmful for one child may be valuable for another. This could include such features as time limits or restricting the use of sycophantic language, as well as restrictions on points, rewards, and other incentives.

Do Encourage Digital Literacy

The American Psychological Association says that AI literacy is an “imperative” skill for children and teens to learn, which should include the responsible use of AI systems, ways to spot misinformation and disinformation, and potential harms such as excessive screen time that detracts from in-person interaction and real-world commitments or unhealthy emotional connections.[12] Platforms and policymakers alike should strike a balance between implementing emotionally engaging chatbots that present information in appealing ways and encouraging safe use with meaningful protections that address hallucinations and misinformation.

While warning labels and disclosures are mostly ineffective, it is worthwhile to educate users, especially children, on how unhealthy relationships with technology can develop, on how to know when information is coming from a reliable source, and that AI systems themselves do not experience emotions.[13] Policymakers can also encourage digital literacy that emphasizes teaching children how to identify trustworthy sources to learn how to effectively spot misinformation.[14]

Compared with adults, children and adolescents are still developing critical thinking and self-regulation skills. Children’s ability to understand long-term consequences and learn deductive reasoning skills are generally not fully developed until around the age of 15 or 16.[15] Digital literacy efforts both at and leading up to these ages should give children the tools to apply their developing critical thinking skills in the context of chatbots, with parental controls present in the meantime to fill the gaps where children may not be cognitively ready to mitigate certain risks to themselves in the ways adults can.

Do Require Paid Content Disclosures

In the United States, the Federal Trade Commission (FTC) requires written disclosures when users post because of a relationship to a brand, such as after having received free products, payment, or discounts, or due to familial ties.[16] This requirement has led to influencers writing clear disclosures on social media posts of terms such as “ad” or “sponsored.” Even before social media, the FTC required similar disclosures for paid results on search engines, with clear visual separation such as shading or borders, in addition to labels such as “sponsored” or “ad” on those paid listings.[17] Policymakers should require disclosures of when chatbot outputs contain paid sponsored content similar to required disclosures on web searches and social media paid posts.

Do Create Clear and Distinct Definitions, Including Carveouts

Like social media, what constitutes “harm” is subjective and can create an avenue for censorship. The “duty of care” or “duty of loyalty” to shield children from harm is language that crops up in several chatbot and social media bills, most notably KOSA, which was included in the Kids Internet and Digital Safety Act (KIDS Act), introduced in March 2026 and passed the House in June 2026, as well as in Sen. Marsha Blackburn’s (R-TN) proposed TRUMP AMERICA AI Act, also introduced in March 2026.[18] Specifically, these online services would have a duty to ensure that their design features prevent and mitigate harm to minors. But this language is vague, which would complicate compliance and likely lead to expensive, frivolous lawsuits.[19] Overly broad definitions, such as duties of care, are an attempt to shield children from harmful content but often result in the overregulation of these technologies. To avoid litigation, online services may overcorrect and make it more difficult for minors, and potentially all users, to access helpful content related to mental health, suicide, eating disorders, sexuality, and more.

“Chatbot” has become a convenient catchall for many AI tools, even though doing so blurs important distinctions. Most state and federal legislation compounds this problem by failing to distinguish between AI companions, or systems specifically designed to simulate emotional connection and human relationships, and general-purpose AI chatbots that do not simulate relationships. For example, while the original Children Harmed by AI Technology Act (CHAT Act), introduced by Sen. Jon Husted (R-OH), claims to focus on AI companions, its language would sweep in ChatGPT, Google’s Gemini, Anthropic’s Claude, and even Amazon’s Echo or Apple’s Siri—none of which are designed to be AI companions.[20] Lumping these two types of chatbots together means legislation aimed at preventing parasocial attachments ends up regulating chatbots for retail platforms or customer service, where the risk of emotional dependency is negligible and would create an undue regulatory burden. As a result, policymakers risk regulating more technology than intended. Thankfully, Sen. Husted’s updated CHAT Act 2.0, introduced in July 2026 and with cosponsor Sen. Andy Kim (D-NJ), rightfully makes these distinctions by creating tiered chatbot categories.[21]

Similarly, Rep. Erin Houchin’s (R-IN) Safeguarding Adolescents From Exploitative BOTs Act (SAFE BOTS Act), introduced in December 2025 and included in the KIDS Act, rightfully carves out retail sites, such as chatbots on the sides of retail sites that give recommendations on what to purchase.[22]

Do Set Clear Standards for Licensed AI Professional Services

While many individuals understand the limitations of general-purpose chatbots relative to licensed human professionals, some have relied on false or misleading information for legal advice, leading them to fire their lawyers, while others have followed chatbot medical advice that contradicted doctors’ recommendations.[23] Pennsylvania announced in May 2026 a lawsuit against Character.AI, an AI companion platform, over alleged false labeling that the chatbot was a licensed mental-health professional.[24]

In order to address this, some bills limit or ban chatbots from equating their responses to those by licensed mental-health professionals. Advocates for this policy state that, since human health care professionals must undergo processes for licensure in a state, chatbots providing that same advice should essentially undergo those same processes.[25]

Illinois and Nevada regulate the use of AI systems, including chatbots, in the delivery of licensed mental or behavioral health services.[26] New York’s proposed S7263 would impose liability for damages caused by a chatbot that impersonates a licensed professional in not just mental-health services but other health and legal services as well.[27] At the federal level, the CHATBOT Act and the People-First Chatbot Act, introduced by Reps. Valerie Foushee (D-NC) and Greg Casar (D-TX) in July 2026, prohibits chatbots from implying possession of a license to practice in the healthcare, insurance, and financial industries.[28]

However, there will likely come a time when certain chatbots, trained on reliable data and reviewed by professionals, will effectively provide the same services as a licensed professional can. In these cases, policymakers should not set higher standards for chatbots than they would for human professionals. But they should set standards commensurate with those that a human professionals must meet to practice in a certain field.

One useful model is Utah’s HB 452, which requires licensed mental-health therapists to participate in the development and review process for mental-health chatbots and that the chatbot’s development is consistent with clinical best practices.[29] It also rightfully says to, “ensure that the output of the mental-health chatbot poses no greater risk to a user than that posed to an individual in therapy with a licensed mental health therapist” via testing.[30]

Congress should also require platforms to treat AI companions that serve roles similar to mandatory reporters—such as teachers, therapists, or nurses—with similar obligations to report abuse. However, policymakers will need very clear definitions, as any chatbot that has “mandatory reporter-like” features would still have obligations. This could be an avenue where more technologies than necessary get swept up in poorly crafted bills. Industry also has a role to play in crafting self-standards for these clarifications.

Do Address Child Suicidal Ideation Conversations

Throughout the United States, there have been multiple documented legal cases of children accessing overly sexual, violent, or suicidal content on chatbots, leading to some children taking their own lives or severely injuring themselves in part due to unhealthy relationships with these technologies.[31] These cases reveal a clear need for chatbot platforms themselves, along with policymakers and parents, to better address the issue of conversations with chatbots delving into topics of extreme physical or mental harm.

Research from Anthropic in 2026 finds that the biggest reason users turn to Claude is for health and wellness.[32] Additionally, a study by Stanford University in 2025 finds that one in eight adolescents use chatbots for mental-health advice.[33] Much of the United States faces mental-health resource droughts: more than half of rural America has a mental-health professional shortage, impacting more than 122 million Americans.[34] Aside from supply-demand issues, there are patient-level barriers to accessing mental-health care, such as stigma and cost, in addition to provider-level barriers such as reimbursement challenges and burnout.[35] As of October 2021, the American Academy of Pediatrics, the American Academy of Child and Adolescent Psychiatry, and the Children’s Hospital Association had declared child and adolescent mental health a national emergency.[36] American children are facing a mental-health crisis, and chatbots could provide an avenue for filling care deserts with widespread free and safe mental-health resources, if utilized safely and effectively.

AI companions are available 24/7 for support, including times when vulnerable or lonely young users may not be able to talk to a real-life friend or family member, or when those trusted adults are ill equipped to handle mental-health conversations. These services provide crucial benefits; however, they also introduce new and serious risks.

To address youth mental health and engagement with chatbots, certain bills require crisis-referral protocols whenever a user engages in a conversation with a chatbot and displays signs of suicidal ideation. California’s SB 243 requires not only operators of companion-chatbot platforms to implement a protocol for addressing suicidal ideation, suicide, or self-harm, including a notification that refers users to crisis-service providers, but also annual reporting on the connection between chatbot use and suicidal ideation.[37] Specifically, operators must annually report to the Office of Suicide Prevention the number of times they issued crisis-referral notifications, protocols put in place to detect and respond to instances of suicidal ideation, and protocols to prohibit chatbot responses about suicidal ideation. Oregon, Georgia, Idaho, Nebraska, New Hampshire, New York, Washington, and Iowa also take similar approaches, as do federal bills such as the CHAT Act and the KIDS Act.[38] While these bills contain other flaws, these specific precautions can help children learn how to access mental-health resources.

Policymakers and developers should combine these precautions with the aforementioned standards for licensed AI mental-health services. For example, a general-purpose chatbot not designed to sensitively and safely handle topics related to suicidal ideation, suicide, or self-harm would face additional restrictions on engaging in these topics compared with a chatbot proven to meet the necessary standards to help children navigate mental-health issues.

Do Discourage Misuse of Fictional Characters Popular With Children

Parasocial relationships are not inherently harmful, as they can play a role in identity formation for children and adolescents. Research suggests that imagining relationships and expressing emotions toward characters or celebrities from a distance can provide a “safe forum” for exploring different aspects of one’s personality.[39] These relationships are not unusual. Many children develop parasocial bonds with traditional media characters, such as Elmo from Sesame Street, or real-world figures such as social media influencers. Adults, too, often engage in parasocial relationships, such as the millions of diehard Taylor Swift fans.[40] In fact, some estimates claim that up to 51 percent of Americans have experienced a parasocial relationship.[41]

But children are less able to distinguish between what is real and what is imaginary compared with adults, and therefore, some children may confuse fictional experiences with real ones. Confusion between reality and imagination is a normal part of child development, but questions remain whether chatbots will exacerbate this confusion and hinder cognitive development.

Chatbots and companions that misuse parasocial relationships, such as by using unauthorized depictions of beloved, trusted characters to create harmful or disturbing content, is an issue policymakers should follow closely to ensure that copyright owners can enforce their intellectual property rights to prevent the misuse of their copyrighted characters.

Do Fund More Research

Policymakers should recognize the diverse ways in which this technology could impact loneliness and social connections. Since the effects of chatbots and AI companions remain under-researched, Congress should prioritize funding studies on how AI companions affect different groups, particularly children.[42] For example, the CHATBOT Act directs the National Science Foundation to study chatbots’ effects on children’s social development, and Rep. Erin Houchin’s (R-IN) AI Warnings and Resources for Education Act (AWARE Act), introduced in September 2025 and included in the KIDS Act, directs the FTC to make public educational resources on responsible use of chatbots.[43] Without sufficient data to understand the full scope of impacts on society, policymakers risk undermining potential benefits by acting too hastily.

Do Encourage Industry Best Practices for Self-Regulation

Policymakers should encourage the AI industry to develop safety standards analogous to those used in other consumer product categories, such as manufacturing equipment and other information technologies.[44]

Meta, for example, has established guidelines for teens interacting with AI characters, banning access unless a parent approves, and OpenAI has established age-prediction measures that adjust the behavior of its models based on inferred user age.[45] Industry coalitions should build on these efforts to develop shared baseline standards, covering areas such as emotional dependency disclosures and escalation protocols for self-harm that apply across platforms. Platforms with characters that serve in “mandatory reporter” roles, or that are designed to provide emotional support, wellness guidance, or mental-health-adjacent functions, should also consult with mental-health and health care professionals to adopt industry-wide safeguards to reduce harm, drawing on established clinical frameworks for recognizing crisis, setting appropriate boundaries, and referring users to qualified human care when the stakes exceed what any AI system should manage alone.[46]

Chatbot Policy “Don’ts”

The following approaches would fail to deliver meaningful protection for youth—and worse, would create collateral damage that restricts beneficial applications of chatbots for users of all ages:

Don’t Require Age Checks for All Users

First, there are many calls at both the state and federal level to enact age-verification regimes on chatbots. Senator Josh Hawley (R-MO) introduced the Guidelines for User Age-verification and Responsible Dialogue Act (GUARD Act) in October 2025, cosponsored by Sens. Richard Blumenthal (D-CT), Katie Britt (R-AK), Mark Warner (D-VA), and Chris Murphy (D-CT), which would require mandatory age verification for AI companion chatbots in order to ban minors from the platforms entirely.[47] Michigan and Oklahoma likewise have their own proposed age-verification rules for chatbots.[48]

Current age-verification approaches range from self-reporting mechanisms that are easy to falsify but minimally invasive to ID checks that are more accurate but also more invasive, requiring users to disclose their real identities to use online services, and less accessible, since not all adults have government IDs.[49] Existing age-verification laws have targeted adult websites and social-media platforms, but these new bills indicate that chatbots are the next target.

Compared with ID checks, AI age estimation is a more-privacy preserving method of age verification. Earlier this year, OpenAI’s ChatGPT announced its age prediction methodology, which includes a “combination of behavioral and account-level signals” such as how long an account has existed and patterns in the times of day each user is active.[50] If the platform estimates that a user is a minor, it places the user in an “under-18 experience,” which contains filters for “harmful” content.[51] If the platform accidentally places a user in the under-18 experience, they must verify their age with a face scan by taking a selfie. Its accuracy varies depending on image quality, age range, gender, and region of birth, but one major age estimation service, Yoti, demonstrates a positive rate for 13- to 17-year-olds at 99.3 percent, and age estimation software continues to improve.[52]

When it comes to chatbots, age-verification requirements often apply only to certain types of services, such as AI companions that engage in sexual or violent conversations. But drawing those lines is not easy; reasonable people may disagree on what content meets certain criteria. Is a conversation about sexual orientation a “sexual conversation”? Is a conversation about a violent historical or current event a “violent conversation”? These blurred lines can result in chatbot services restricting too much or too little content. They also take the decision out of the hands of parents who may be better able to decide what AI chatbots are appropriate for their children based on their own knowledge about their children’s maturity, level of adult supervision, and intended use.

Don’t Restrict Content

The Kids Online Safety Act (KOSA), now enmeshed in the KIDS Act, also includes language meant to shield children from harmful content—though again, the definition of “harmful” is up for interpretation.[53] These content-related requirements either give governments the power to determine what content is harmful to children—a likely First Amendment violation—or force the businesses that offer chatbot services to make those decisions, potentially opening themselves up to legal consequences if they do not correctly identify harmful content.

Some legislation, such as the CHATBOT Act would collect logs of children’s conversations with chatbots.[54] However, many young people turn to chatbots to have private conversations about certain topics—such as LGBTQ+ identities, mental or physical health issues, relationships, and more—that they may feel unsafe or uncomfortable having in real life. While parents may often have a legitimate safety interest in seeing their children’s chatbot logs, requiring platforms to collect this data carries serious risks, especially for children whose parents do not have their best interests in mind or older adolescents who need greater autonomy from their parents and room to explore their identities.

Finally, policymakers should avoid banning chatbots from providing financial, legal, or medical information. Users are free to search for that kind of information on the Internet if they want. They should be free to ask a chatbot to synthesize the information for them, too.

Don’t Perpetuate the “Addiction” Narrative

As is the case with social media, many policymakers have considered banning or restricting chatbots over “addiction” concerns. These fears mirror decades of similar concerns across all sorts of technologies, from television to video games to social media.[55] However, “addiction” is not a scientifically or medically accurate term for chatbot use: Current research does not support that narrative, and policymakers should therefore avoid this trope.[56]

Policymakers who rely on the addiction narrative often compare the risks chatbots pose to children to those posed by alcohol or cigarettes. But there is indisputable scientific research showing clear linkages between cigarette use and increased risks of cancer and lung disease, and the negative impact on health of underaged drinking.[57] In contrast, the research does not support the claim that social media, smartphones, or chatbots cause health harms of a similar magnitude.[58]

Moreover, alcohol and tobacco products are always harmful to children, regardless of brand or flavor, because they all contain the same dangerous chemical compounds.[59] Chatbots, by contrast, are not all the same, and treating all platforms as such reflects, at best, sloppy reasoning and, at worst, a misleading argument.

Don’t Rely on Warning Labels to Educate Users on Harms

Following a 2024 U.S. Surgeon General report that argues that social media poses such a threat to children’s mental health that the platforms should come with warning labels like cigarettes and alcohol do, policymakers have called for similar measures for chatbots.[60] Almost all proposed chatbot bills include language requiring a disclosure that the user is not chatting with a human or warning of excess time spent on the platform.[61]

Mandating warning labels is a flawed approach that shows concern while not actually providing any meaningful protection. Most users ignore warning labels, and children don’t always understand disclosures and the warning labels they read, nor do many adults.[62] If the goal is to educate users, and particularly children, on the potential risks involved in using chatbots, warning labels are insufficient, and will become similar to cookie banners that the vast majority of users click through without reading or understanding—a meaningless box for platforms to check, wasting users’ time and platforms’ money.[63]

Don’t Ban Targeted Advertising

Targeted advertising delivers custom ads to users based on certain demographics, interests, or location. Much of the Internet relies on targeted advertising as a source of revenue instead of charging users a fee for services. Taking this revenue away would lead to a greater lack of resources, especially free or low-cost resources, for children, which would be especially detrimental to lower-income households.[64]

Moreover, many concerns about targeted advertising stem from a misunderstanding of how targeted ads work.[65] Online services collect information about their users, but in most cases, they do not sell that personally identifiable information to third parties for advertising. Instead, they sell the opportunity to advertise to a certain demographic. Companies that want to advertise to a younger audience, then, are not purchasing young people’s data; they are purchasing the digital ad space that will show up on young people’s screens. This process is anonymous and much more privacy protective than critics purport.[66]

Unfortunately, the CHATBOT Act would ban targeted advertising on chatbots, and the Youth AI Privacy Act, introduced by Sen. Ed Markey (D-MA) in March 2026, would ban all forms of advertising on chatbots, not just targeted ads, which would have an even greater negative impact on these services.[67] Sen. Markey’s Youth AI Privacy Act argues that advertisements can be “woven directly into responses or may subtly steer users toward certain products based on how the model was trained.”[68] A more targeted approach is Utah’s HB 452, which requires advertising disclosures specifically on mental-health chatbots.[69]

Don’t Ban Chatbots for All Minors

Multiple states have passed or considered legislation that would restrict children under a certain age from accessing chatbots without parental consent, or at all. At the federal level, the GUARD Act would ban AI companions entirely for youth under 18, and Sen. Jon Husted (R-OH) introduced the Children Harmed by AI Technology Act (CHAT Act), which would ban youth from using AI companions unless their parent or guardian registers the account.[70] However, the CHAT Act 2.0 rightfully removes this provision by replacing it with self-reporting of age, with requirements for parental contact information and protective settings in place for minors’ accounts.[71] In California, the proposed AB 1064 would have banned companion chatbots for users under 18, but it was rightfully vetoed by Governor Gavin Newsom.[72]

Banning chatbots altogether would cut youth off from the 24/7 support that safe chatbots and AI companions can provide.[73] Bans would also require these services to implement age-verification mechanisms for all their users to determine which ones to cut off from services, which presents privacy risks for adults. Different verification methods come with varying trade-offs, including requiring users to disclose their real identities in order to use these services.[74]

Bans pose many of the same First Amendment issues as content restrictions do, as they can result in large swaths of speech becoming inaccessible to entire categories of users—not only young users but also adults unable to verify their age or those unwilling to sacrifice their anonymity to do so.[75] There are also legitimate questions about the efficacy of bans where they have been tried, as children often circumvent them.[76]

Don’t Scapegoat Sycophancy

One common critique of chatbots is that they utilize sycophantic language: telling the user only what they want to hear and affirming anything the user says. They argue a sycophantic chatbot model might echo a user’s false beliefs, even when their beliefs are incorrect, which could include encouraging harmful behaviors or fostering pre-existing delusions.[77] But calls to take this ability out of chatbots nullifies why many users turn to chatbots in the first place.

Indeed, a study published in Science proves that users prefer sycophantic AI models, even when the models give bad advice.[78] This proves that affirming language paired with safe systems is what users are most likely to enjoy using. But New York’s proposed S9051, would ban chatbots that contain “unsafe features,” including “outputs that prioritize flattery or sycophancy.”[79] California’s proposed SB 1119 would ban platforms from engaging in responses that are “excessively sycophantic.”[80] Similarly, the GUARD Act would ban minors from using chatbots that simulate friendship or companionship.

“Sycophancy” is an overly broad term and difficult to define in this context, as users could easily interpret responses differently depending on their preferences. Chatbot responses that use “please” and “thank you” could be viewed as polite by one user and sycophantic by another.

“Sycophancy” and “addiction” are the new chatbot boogeymen just as algorithms are to social media, and these arguments are just as meritless.[81] For example, in 2023, Meta offered Instagram and Facebook users the ability to switch from algorithmic feeds to the traditional chronologic feeds, as many users blamed (and continue to scapegoat) algorithms for their negative experiences on social media.[82] As it turns out, many users preferred the curated and personalized algorithmic feed over a chronologic one, proving that users often prefer online experiences that are more tailored and personalized.

Similarly, some chatbot platforms allow users to tailor the level of politeness or responsiveness according to their personal preference. ChatGPT users can choose from “personalities” such as “candid,” which gives “no-nonsense, plain-spoken input” to “quirky,” which provides jokes or “imaginative framing.”[83]

Importantly, not all large language models (LLMs) are sycophantic, so calls to ban or restrict chatbots or other LLMs because of this feature may sweep in more technologies than are necessary. Not every AI response on a platform will be sycophantic, even on platforms that allow such responses. For example, research from Anthropic finds that only 9 percent of its chatbot’s answers are sycophantic.[84] Platforms could also give users control over sycophantic language through parental settings, letting them decide for themselves how they want their children to engage, rather than banning sycophancy altogether.

Don’t Regulate Time Limits

Government-set time caps are a blunt mandate that undermines parental authority, creates new compliance burdens, and expands reliance on age verification. Time-limit policies stem from a flawed but prevailing narrative that chatbots are inherently addictive, particularly for young users, and therefore, youth need strict limits to their time online.[85]

For example, the CHATBOT Act would require the creation of family accounts to oversee under-13-year-olds’ chatbot usage, which would include time and memory limits. While encouraging parental oversight and calling for improved parental controls that allow for customization for these features is welcome, this would attempt to legislate what most platforms already have in place.

Just like social media time limits, chatbot time limits are likely to face First Amendment lawsuits because they restrict access to speech online.[86] Moreover, as with bans, children can evade time limits simply by creating multiple accounts to add time to their session or using their parents’ or friends’ accounts. Some bills, such as California’s SB 243, requires periodic reminders every three hours for minors to take breaks—similar to many streaming platforms’ reminders asking viewers whether they are still watching after a certain amount of time has passed.[87] While this regulation avoids hard time cutoffs, chatbots should have the same choice as streaming services do as to whether to enable these reminders and how often they should occur. Mandating reminders at certain intervals for only certain types of platforms is inconsistent and ineffective government policy.

Conclusion

Protecting children from the real potential harms of chatbots does not require banning or severely restricting a technology that holds enormous promise for education, mental-health support, and personal development. Policymakers should resist the temptation to recycle failed social media regulations or react to moral panic. Instead, thoughtful regulation should center on clear definitions, meaningful parental controls, industry standards, and targeted safeguards that address documented harms without cutting children off from these beneficial tools.

Endnotes