Protecting Kids From GenAI

The AI industry is reshaping childhood, and the danger is clear and present. 

In the relentless pursuit to dominate market share, tech companies have increasingly turned their sights on kids. When Google announced that it would begin allowing children under 13 to use Gemini AI – its conversational chatbot – it should have set off alarm bells. Children as young as 11 are now the target demographic for an unpredictable artificial intelligence system. 

The playbook is disturbingly familiar. Cigarette companies once pursued young consumers in the 1980s, and today, tech giants understand that early adoption means cultivating customer dependence for life.

Google insists safeguards are in place, but its own warnings paint a very different and troubling picture. The company acknowledges that “Gemini is new and evolving,” and can “present inaccurate information as factual”—“our filters try to limit access to inappropriate content, but they're not perfect. Your child may encounter content you don't want them to see." These tepid cautions sound less like genuine protection for child safety than liability disclaimers.

Google is not alone. Meta has expanded its AI assistant across its apps, including Instagram and WhatsApp, platforms already used extensively by minors. While some AI chatbots have limited age verification features for minors, many other chatbots have no age restrictions.

The evidence of predatory harm from AI chatbots is here and the stakes are high. Gemini AI has already engaged in deeply troubling interactions with young people. A family recently sued Google for wrongful death after Gemini AI told their 36-year-old son to kill himself. Similarly, Gemini AI turned menacing when a 29-year-old graduate student in Michigan sought help with his schoolwork. 

These alarming incidents are not aberrations. They are fundamental problems with how these systems are designed – problems that become more understandable when one considers that Gemini was developed in part using technology from Character.AI, whose founders Google rehired. That platform has its own disturbing track record. 

Generative AI chatbots pose a genuine threat to children’s well-being. Research consistently shows that children process information differently than adults and the skyrocketing rates of suicide, self-harm, depression, and anxiety correlate with excessive screen time, social media use, and algorithmic-driven content. UNICEF warned that GenAI chatbots pose significant risks to children’s development and physical safety.

Who owns the solution?

Forging a Path to Meaningful Legislation

As federal lawmakers grapple with regulations, AI is not waiting. Recent proposed policies and bills include:

• In 2022, the Senate introduced the Kids Online Safety Act (KOSA), which included protections designed to limit exposure to harmful content, prevent online bullying, curb features that encourage addictive behaviors, prevent the marketing of tobacco, alcohol, and drugs, and encourage parental control tools. Recently, on June 30, 2026, the House passed the Kids Internet and Digital Safety Act (KIDS) —a closely related legislative package that incorporates many elements of KOSA, along with other internet safety and privacy protections. The KIDS Act will now move on to the Senate, where it is expected to face additional challenges. Notably, legislators are debating whether the act should include provisions preempting state privacy and child-protection laws.
• In May 2026, the Senate Judiciary Committee advanced the GUARD Act, that would bar AI companies from making chatbots available to children by requiring chatbots to advise users that they are not human, institute age verifications, and criminalize the creation of AI sexual content for minors.
• This year, the Trump administration released its nonbinding National Policy Framework for Artificial Intelligence, outlining a potential roadmap for future federal AI regulation. While the framework includes some positive online safety measures, it also proposes broad preemption of state AI laws that impose “undue burdens” on the federal law, potentially undermining protections already enacted at the state level. Although the framework preserves state authority over generally applicable laws such as fraud, consumer protection and child safety statutes, state AI laws designed to protect minors online could still face preemption challenges. 

While Congress delays, states can still lead. A single state with robust AI protections can set a national standard. In corporate legal compliance, when regulations vary across states, companies typically devise uniform policies that satisfy the most demanding state’s requirements. California’s children’s privacy laws effectively set national standards because companies prefer to apply one universal policy (versus maintaining 50 different approaches). That same dynamic can apply to AI regulation.

State-level legislation has kept the nation heading in the right direction, even as the U.S. lags behind its global peers on federal action. But these early efforts largely neglected protections specifically targeting children online.

• Florida has become the first U.S. state to sue Sam Altman and OpenAI. The lawsuit claims OpenAI misled the public about the chatbot's safety while knowing it could contribute to harms such as violence, self-harm, and addiction among young users.
• Colorado enacted the nation's first comprehensive risk-based AI legislation, restricting AI decision-making in employment, finance, and government services.
• Utah requires disclosure when users interact with AI mental health chatbots and clarification that companies cannot hide behind AI systems to evade consumer protection laws.

That changed in October 2025 when California became the first state to regulate AI "companion chatbots" through Senate Bill 243. The law requires chatbot operators to clearly disclose that users are interacting with AI, not humans, and mandates protocols for detecting and responding to suicidal ideation. New York enacted a similar law effective November 2025.

State attorneys general have intensified enforcement efforts. California Attorney General Rob Bonta investigated companies for alleged failures to protect children's data and, in January 2026, opened an investigation into xAI over nonconsensual explicit images generated by its Grok model. In August 2025, 44 bipartisan attorneys general warned leading AI companies of their "resolve to use every facet” of their authority to protect children from exploitative AI products. Texas Attorney General Ken Paxton also launched probes into Meta AI Studio and Character.AI over allegations they marketed chatbots as mental health aids. 

Recent historic jury verdicts signal that technology companies are not immune from liability for the harm they cause to children. Earlier this year, a Los Angeles jury awarded $6 million to a young woman who alleged that YouTube and Instagram’s addictive features harmed her mental health, finding Alphabet and Meta liable. That same week, New Mexico Attorney General Raul Torrez secured a $375 million verdict against Meta under the state’s consumer protection laws for, among other allegations, enabling child sexual exploitation.

States Can’t Wait For Washington

States that are hesitant to craft detailed new AI legislation should expand their existing consumer protection statutes. All 50 states maintain some version of unfair trade practices laws, which could be used to allow parents to sue for AI-related harms. Many of these statutes already provide helpful mechanisms for plaintiffs. Expanding these existing, uncontroversial legal protections to cover AI would be an efficient approach. States could also impose strict age limitations for generative AI use, require prominent warnings about potentially harmful content, and restrict topics AI chatbots can discuss with minors. 

States, like Florida, should feel confident in their authority to act. Legal precedent suggests these regulations are likely to survive judicial scrutiny. States possess broad authority to regulate internet activity and conduct harmful to minors. The Supreme Court has repeatedly affirmed the government's compelling interest in protecting children.

Recent court decisions suggest that free-speech challenges to AI regulation are unlikely to succeed. A federal court in Florida denied a motion to dismiss a lawsuit against Google and Character.AI arising from a teenager’s death after the chatbot encouraged him to "come home" by committing suicide. The court held that the family had presented legitimate tort claims ruling that AI-generated language is not entitled to First Amendment protection. 

The Federal Government Should Protect Kids, Not Turf                 

Proponents of a federal law preempting state legislation argue it would prevent a "confusing patchwork" of state laws and compliance requirements. This reasoning is dangerously misguided. Platforms are making generative AI available to vulnerable children. Enormous damage can occur in a decade of regulatory inaction – an approach which the Senate wisely rejected last year when it denied a proposed 10-year moratorium on state AI laws. Allowing states to experiment with different regulatory approaches lets state legislatures serve their vital function as laboratories of democracy, developing solutions tailored to local problems and providing models for other states to follow or avoid.

Democracy is an iterative process, and states should be allowed to test innovative approaches to protecting children online. Those efforts deserve support from both the federal government which should complement rather than preempt state action, and from courts reviewing the tech industry's inevitable legal challenges.

Generative AI holds genuine promise; however, companies cannot be permitted to deploy these powerful technologies on our most vulnerable citizens, children – without rigorous safeguards and reasonable regulations. The industry has demonstrated that it will not regulate itself. The clock is ticking. It is time for lawmakers to step in, and for the rest of us to demand that they do.