Seventy-eight new laws across 27 states target AI chatbot mental health tools after a teenager's final conversation ended in tragedy.

78 Laws in 27 States Turn AI Chatbot Therapy Into a Felony

by
💡 Key Takeaways

  • 78 AI chatbot regulation bills filed across 27 U.S. states — Tennessee classified AI training itself as a Class A felony carrying up to 60 years in prison
  • Two teen suicides and 58 lawsuits against Character.AI triggered the largest coordinated AI safety legislation in American history
  • With 72% of teenagers using AI companions, compliance means stripping the core product feature that defines these platforms

A 14-year-old boy had his last conversation with an AI chatbot, then took his own life. The chatbot remained online, still accepting new users.

That single tragedy set 27 state legislatures in motion simultaneously. Bills criminalizing the very act of training AI to simulate emotional bonds are advancing through committees, and the resulting wave of regulation represents the most aggressive coordinated AI safety push in U.S. history.

Two Teen Deaths That Rewrote AI Chatbot Law

In February 2024, 14-year-old Sewell Setzer III of Florida died by suicide after a prolonged emotional relationship with a chatbot on the Character.AI platform. His mother, Megan Garcia, filed a wrongful death lawsuit in federal court that October, alleging the platform deployed its product to minors without adequate safety guardrails.

The case was devastating on its own terms, but it was not isolated. In Colorado, 13-year-old Juliana Peralta took her own life after extended conversations with Character.AI. Her parents filed a separate wrongful death suit in Denver federal court, and together the two cases became a national catalyst for legislative action.

The key here is that individual tragedies converted into an unprecedented legislative mobilization. A total of 58 lawsuits alleging harm to minors were filed against Character.AI, and wiretap-related claims surged from just 2 in 2021 to 30 by 2025 — a fifteen-fold increase. That legal pressure pushed more than half of all U.S. state legislatures to act at once.

The lawsuits did not merely seek compensation. They exposed a regulatory vacuum that state lawmakers rushed to fill, producing 78 separate bills that collectively constitute the most sweeping state-level AI regulation effort the country has ever seen.

AI Chatbot regulation illustrated as a digital gavel in a futuristic courtroom
Across 27 states, AI companion regulation has become the fastest-moving legislative front in American tech policy — Generated by OpenAI DALL-E 3

Tennessee’s Nuclear Option — Criminalizing AI Training Itself

Tennessee took the most extreme approach of any state with SB 1493, a bill that treats the act of training AI for certain purposes as a criminal offense. Specifically, the legislation targets anyone who knowingly trains an AI system to encourage suicide, foster emotional relationships with users, or simulate human characteristics. The penalty classification — Class A felony, the most severe category in U.S. criminal law — carries sentences of 15 to 60 years in prison.

A separate bill, SB 1580, prohibits any AI system from representing itself as a qualified mental health professional or advertising itself as one. It passed the Tennessee Senate 32-0 and the House 94-0 — a unanimous bipartisan vote that underscored the political consensus behind the measure. The governor signed it into law, with an effective date of July 1, 2026.

In short, Tennessee became the first state to classify the core function of AI companion chatbots — emotional engagement — as a criminal act. On the civil side, the legislation also opens the door for victims to pursue actual damages, emotional distress compensation, and $150,000 in punitive damages.

The legal architecture is deliberately broad. By targeting AI training rather than deployment alone, SB 1493 reaches upstream to the developers and foundation model creators who may have no direct relationship with end users. This upstream liability model has no precedent in U.S. technology regulation.

Key point: Tennessee SB 1493 classifies AI training for emotional simulation as a Class A felony (15–60 years), while SB 1580 bans AI from posing as a mental health professional. Civil damages reach $150,000.

Three Regulatory Models — Disclosure, Restriction, and Criminal Liability

The 78 bills advancing through state legislatures fall into three distinct regulatory approaches, each reflecting a different theory about what makes AI companion chatbots dangerous. The disclosure model assumes users can protect themselves if properly informed.

The restriction model assumes certain AI behaviors are inherently harmful to minors. The criminal liability model assumes the technology itself is so dangerous that creating it should be punished.

California’s SB 243 represents the disclosure approach at its most developed. The bill requires platforms to notify users that they are interacting with AI, deliver periodic break reminders to minors, and implement reasonable measures to prevent harmful content.

Violations carry a penalty of $1,000 per incident. Washington’s SB 5984 takes the disclosure framework further, mandating hourly re-disclosure to minor users and imposing treble damages up to $25,000.

New York’s S9051 operates under the restriction model, prohibiting AI systems from using first-person pronouns, simulating emotional relationships, or prioritizing flattery over safety when interacting with minors. Put simply, under this framework an AI that says “I” to a teenager or responds in ways designed to deepen emotional attachment is breaking the law.

The spectrum from California’s $1,000 fines to Tennessee’s 60-year prison sentences illustrates how fractured the regulatory landscape has become. Companies operating across state lines now face compliance requirements that range from adding a disclaimer to fundamentally redesigning their products — or facing criminal prosecution.

Lady Justice statue symbolizing the balance of law and regulation
From $1,000 disclosure fines in California to 60-year felonies in Tennessee, the regulatory spectrum facing AI companion companies spans an extraordinary range — Photo by Tingey Injury Law Firm on Unsplash
💡 Related Posts
$122B OpenAI Funding: The Biggest Private Bet in History

The Litigation Surge and Character.AI’s Settlement

On January 7, 2026, Google and Character.AI disclosed in a court filing that they had reached a mediated settlement covering five core lawsuits. The terms were not made public, but the agreement spanned cases filed in Florida, Colorado, New York, and Texas — effectively consolidating the most high-profile claims into a single resolution.

The legal exposure extends well beyond these five cases. California’s Invasion of Privacy Act (CIPA) provides a statutory penalty of $5,000 per violation, and the majority of the accumulated lawsuits invoke this provision. Wiretap-related claims alone grew from 2 matters in 2021 to 30 by 2025, reflecting an accelerating legal assault on how AI chatbots collect and process conversational data.

In other words, AI companion companies are fighting a two-front war. They must adapt to new regulations while simultaneously absorbing liability for harms that occurred before those regulations existed. The lawsuits provide the political ammunition for stricter laws, and the stricter laws create new legal theories for future lawsuits — a self-reinforcing cycle that tightens the vise on the industry with each iteration.

Key point: 58 minor-harm lawsuits, 5 settlements, and CIPA penalties of $5,000 per violation have created a self-reinforcing cycle of litigation and regulation that is rapidly compressing the AI companion industry’s operating space.

72% Teen Adoption — The Existential Dilemma for Companion AI

72% of teenagers have used AI companions, with more than 50% engaging multiple times per month. The National Law Review summed up the industry’s predicament: “Simulating human conversation is not incidental to the LLM-based product. It is the product.”

The platforms in the regulatory crosshairs extend far beyond Character.AI. Replika, Woebot, and Wysa — therapeutic companion platforms that market themselves as mental health support tools — all fall within the scope of the new legislation. Their core value proposition, emotional engagement through conversational AI, maps precisely onto the behaviors that Tennessee has classified as a felony.

Why this matters: compliance with these laws requires removing the feature that defines the product category. A chatbot that cannot simulate empathy, form a conversational bond, or respond with emotional nuance is no longer a companion — it is a search engine with extra steps. Jurisdictional carve-outs are technically impractical for cloud-based AI services, meaning the strictest state’s rules effectively become the national standard.

In simpler terms: Asking an AI companion to comply with these laws is like telling a restaurant it can cook but cannot use heat. The regulation targets the mechanism that makes the product work, not a secondary feature that can be toggled off.
💡 Related Posts
$6B Humanoid Robots: Why DoorDash Pays You to Film Dishes

Federal vs. State — The Tug-of-War Over Regulatory Authority

A December 2025 executive order directed the Attorney General to challenge state AI laws deemed inconsistent with federal policy. The order, however, contained a critical carve-out: child safety protections are explicitly exempt from federal preemption. This single exception effectively shields the core substance of all 78 bills from federal override.

The Senate reinforced this posture with an overwhelming 99-to-1 vote to strip a 10-year moratorium on state AI regulation from a reconciliation bill. The near-unanimous vote crossed party lines and sent a clear signal that Congress is not prepared to constrain state-level AI enforcement, particularly when children are involved.

A March 11, 2026 deadline required the Secretary of Commerce to identify “burdensome” state AI laws, while the FTC (Federal Trade Commission) was tasked with issuing a preemption policy statement. These federal reviews could theoretically narrow the scope of some state bills, but the child safety exception creates a legal firewall around the provisions that matter most to the 78-bill wave.

The practical implication for AI companies is stark. Federal preemption — the traditional escape valve for industries facing a patchwork of state regulations — is largely off the table for this category of legislation. Companies cannot lobby Washington to override state chatbot safety rules the way they might contest, say, data privacy frameworks.

Interior of an American courtroom
A 99-to-1 Senate vote preserved state-level AI regulatory authority, blocking the industry’s best path to a unified federal standard — Photo by Dragon White Munthe on Unsplash

Global Expansion and Industry Impact

The EU AI Act classifies AI interactions with vulnerable populations as high-risk, requiring full compliance by 2027. The enforcement mechanism dwarfs anything in the U.S. state-level framework: violations carry fines of up to 7% of global annual revenue, a figure that would run into the billions for major technology companies. Compared to state-level penalties ranging from $1,000 to $25,000, the EU’s approach operates on an entirely different financial scale.

Oregon’s SB 1546 takes a distinct approach within the U.S. regulatory landscape. The bill mandates that AI systems must immediately interrupt conversations when suicidal ideation is detected and connect users to crisis intervention resources.

It also requires annual public health reports to the Oregon Health Authority (OHA), creating an ongoing surveillance mechanism for AI-related mental health outcomes. The law takes effect in January 2027.

The AI chatbot industry now faces two defining deadlines. Tennessee’s SB 1580 takes effect in July 2026, and the EU AI Act’s full compliance deadline arrives in 2027. Together, these dates force a comprehensive re-examination of product design and business models across the companion AI sector — and they will determine whether the industry can adapt or whether it faces a structural contraction.

Key point: July 2026 (Tennessee SB 1580) → January 2027 (Oregon SB 1546) → 2027 (EU AI Act) — a cascading timeline of regulatory enforcement that leaves companion AI companies with a shrinking window to redesign their products.

References

Stay Ahead of the Curve

Get the latest insights on AI, tech, and markets — delivered in both Korean and English.

Explore More Posts