When looking to regulate emerging technologies, the popular instinct has been to draft these rules from the top down — starting with the most restrictive option. As courts have struck down these proposals, policymakers move on to the next most restrictive idea, repeating the pattern over and over. This strategy has failed in the past, and the outcome will be no different for regulating artificial intelligence (AI). But Congress has an opportunity to part ways with their historically ineffective “top-down” philosophy to regulating the tech industry. By starting with clear, actionable rules that address documented harms, policymakers can better protect kids’ online from the bottom up.
Representative Erin Houchin’s SAFE BOTs Act and Senator Josh Hawley’s GUARD Act embody the conflicting visions for what Congress should do to address growing concerns about chatbots and minors. The SAFE BOTs Act sets rules for how chatbots should interact with users, whereas the GUARD Act sets out to determine who can use them.
The GUARD Act would explicitly require every user of an “AI companion chatbot” to verify their age using sensitive information linked to their real-world identity, like a government-issued ID or personal banking data. If a user is classified as a minor, they are prohibited from using AI companions outright. Minor users, while the focus of the measure, would not be the only group affected. Users of all ages would be subject to privacy-invasive age verification just to use AI tools like ChatGPT or Claude. In violation of their first amendment rights, adults unwilling to divulge their personal information would be restricted from accessing general-purpose AI systems.
Such a broad age verification mandate is likely to be found unconstitutional. As Jennifer Huddleston and Juan Londoño from the Cato Institute explain, the GUARD Act restricts minors’ access to large quantities of protected speech based on “harms of AI that are not sufficiently proven,” and do not “establish adequate government interest.” Early attempts to restrict access to violent video games were struck down for similar reasons. These provisions have doomed federal kids’ safety laws before, as policymakers are naturally unwilling to pass a law that will inevitably be voided by the courts. A previous version of the Kids Online Safety Act cleared the Senate only to stall in the House under the weight of the age verification provision’s free speech concerns. Since proposals with age verification restrictions for AI systems are likely to meet a similar fate, they are an unrealistic vessel for keeping kids safe online and will ultimately fail to protect users as intended.
Instead of banning entire product categories or age groups, policymakers should iteratively draft targeted protections based on demonstrated harms. This would allow guardrails to emerge from the bottom up as AI systems progress and expand if needed. The SAFE BOTs Act is emblematic of this more prudent, measured approach.
Under this proposal, chatbots must disclose that they are an AI system, not a person, at the start of a minor’s first interaction and repeat that disclosure whenever the user asks. If a minor user discusses topics related to suicide or suicidal ideation, the chatbot must surface resources to contact suicide and crisis intervention hotlines. After three continuous hours of interaction, the chatbot must prompt the user to take a break. Similar requirements have already been signed into law in California and New York. In contrast to the numerous state laws mandating age verification, these laws have not been challenged in court.
Chatbot companies would also be prohibited from impersonating a licensed professional such as a doctor, therapist, lawyer, or financial advisor. The proposal does not restrict what chatbots are allowed to discuss — a user can still ask for help filing their taxes or building a budget — but the chatbot can’t claim to be a CPA. These standards are useful to encourage users to verify the accuracy of a chatbot’s suggestions and reach out to a professional if needed,without removing the value that AI tools can provide to making financial or legal resources more accessible.
To be clear, the SAFE BOTs Act is by no means perfect. Disclosure requirements or prohibitions on falsely claiming to be a licensed professional can be sensible guardrails for users of all ages — avoiding the need for age verification. A chatbot can share crisis response resources to all users seemingly in distress, regardless of if the user is 13 or 35. However, by requiring providers to maintain content policies specific to minors, Houchin’s bill may inadvertently pressure companies to verify users’ ages — reintroducing the same constitutional concerns that plague the GUARD Act.
SAFE BOTs requires that chatbot providers must develop and maintain “reasonable policies” addressing minor users’ exposure to sexual content, gambling, illegal drugs, tobacco, or alcohol. Though this provision is ostensibly mundane, the lack of specificity regarding what constitutes “reasonable” efforts can encourage platforms to comply with the strictest interpretation of the law to avoid costly accidental non-compliance. The race toward stricter interpretations eventually lands at content filtering, which requires distinguishing minors from adults, which ultimately invites age verification.
The SAFE BOTs Act demonstrates that Congress can write targeted, durable rules for AI systems grounded in existing consumer protection principles. Yet the bill’s content provisions show how easily a thoughtful bottom-up approach can drift toward the same top-down instincts it sought to avoid. Congress has a real opportunity to build an AI regulatory framework that protects users without repeating the constitutional failures of the past. Seizing it demands commitment to a bottom-up regulatory philosophy, not just good intentions.
