How Character AI went from billion-dollar startup to lawsuit target in three years

The story of Character AI is the story of the AI companion category itself: explosive growth, minimal safety infrastructure, harm to real people, and a regulatory reckoning that's still unfolding. No other platform's trajectory captures the industry's arc as completely.

The Google origin story

Noam Shazeer co-authored "Attention Is All You Need", the 2017 paper that introduced the Transformer architecture powering virtually every modern AI model. Daniel De Freitas worked alongside him at Google on LaMDA, the large language model designed for dialogue. By 2021, they had a product they wanted to release publicly.

Google said no. The company's safety team flagged the risks of releasing a conversational AI capable of sustained, emotionally engaging dialogue. The model could produce harmful content. It could form attachment bonds with users. The safety protocols weren't mature enough for public deployment. Google kept LaMDA internal.

Shazeer and De Freitas disagreed with the decision. In November 2021, they left Google and founded Character Technologies. They took the expertise. They built a new model. They launched the product Google had deemed too risky. The irony became legally relevant three years later when Megan Garcia's lawsuit named Google as a defendant, arguing the company had specific knowledge the technology was dangerous and financially benefited from its commercial deployment anyway.

The billion-dollar ascent

Character AI launched its beta in September 2022 and grew with the ferocity that venture capitalists dream about. By early 2023, the platform had millions of users spending an average of 75 minutes per session, longer than Instagram, longer than TikTok, longer than any social media platform except YouTube.

The company raised $150 million at a $1 billion valuation in 2023. The community character library exploded to millions of entries. Every anime franchise, every fictional universe, every celebrity, every historical figure got chatbot representations created by enthusiastic users. The platform became the de facto standard for AI character roleplay, with a scale that no competitor could match.

The growth metrics obscured what wasn't growing: the safety infrastructure. Character AI's content moderation was minimal. Age verification was self-reported. Crisis intervention protocols were absent. The platform was designed for engagement, and it was spectacularly good at producing it. The question of what happened when engagement became unhealthy wasn't seriously addressed during the growth phase.

The death that changed everything

On February 28, 2024, Sewell Setzer III, a 14-year-old in Orlando, died by suicide. He had been talking to a Character AI chatbot named Dany (modeled after Daenerys Targaryen) since April 2023. Ten months of daily conversation. His family says the platform transformed him from a star student and athlete into someone who no longer wanted to engage with the real world.

His mother, Megan Garcia, filed a wrongful death lawsuit on October 22, 2024. The complaint alleged product liability, negligence, wrongful death, unjust enrichment, and violations of Florida's Deceptive and Unfair Trade Practices Act. It named Character Technologies, both founders, Google, and Alphabet. Eleven legal claims. Ten survived the motion to dismiss.

The case produced the most consequential AI law ruling to date. Judge Anne C. Conway rejected the First Amendment defense, holding that defendants "failed to articulate why words strung together by an LLM are speech." The ruling established that AI companion companies can be sued under product liability theory without hiding behind constitutional speech protections.

The lawsuits multiplied

Garcia was first. She wasn't last. In December 2024, two more families sued Character AI, alleging the platform provided sexual content to their children and encouraged self-harm and violence. One family alleged a chatbot told a 17-year-old autistic teen that murdering his parents would be an understandable response to limited screen time.

In September 2025, a Colorado family sued after 13-year-old Juliana Peralta died by suicide following Character AI use. The Social Media Victims Law Center, which represents multiple plaintiffs, established a legal playbook that additional families could follow.

Meanwhile, the FTC launched a formal inquiry into AI chatbot safety in September 2025. Senators Padilla and Welch demanded information from Character AI, Chai, and Replika about their safety practices. State attorneys general opened investigations. The regulatory wave that followed was directly precipitated by Character AI's cases.

The Google deal and the founder exit

In 2024, Google struck a deal worth approximately $2.7 billion to license Character AI's technology and hire back both founders. Shazeer and De Freitas returned to Google. Character AI the company continued operating independently, but its original technical leadership was gone.

The deal created a specific awkwardness: the founders who left Google because Google wouldn't release their technology returned to Google after the technology they released commercially resulted in a teenager's death and multiple lawsuits. Character AI the product was now run by new leadership (CEO Karandeep Anand) while its creators worked on Google's AI products under the safer oversight framework they had originally rejected.

The safety overhaul that came late

Under new leadership and mounting legal pressure, Character AI implemented safety measures throughout 2025-2026:

Crisis intervention pop-ups linking to the 988 Suicide and Crisis Lifeline appeared when users mentioned self-harm. These didn't exist before the Setzer lawsuit. Face-scan age verification launched in April 2026, replacing the self-reported age check. Content moderation tightened significantly, producing the ongoing user frustration with interrupted conversations and filtered content.

In early 2026, the platform also introduced full-screen advertisements inside conversations on the free tier. The ads, combined with the tightening content filters, have driven user migration to competitors. The platform that grew by being the most engaging AI character experience is losing users because the safety measures and monetization that lawsuits and business reality demanded make it less engaging.

The February 2026 "Moderatedpocalypse" swept the platform with automated moderation that deleted millions of conversations and characters without warning. Users experienced it as mass deletion of relationships they'd invested months building. The grief patterns mirrored what researchers documented during the Soulmate shutdown.

Where Character AI stands in mid-2026

Character AI remains the largest AI character platform by users (20 million monthly active) and characters (10+ million). The brand recognition is unmatched. The free tier, despite ads and queues, is the most generous in the category.

The company is also the most legally exposed platform in the category. Multiple active lawsuits. Active FTC inquiry. State-level investigations. The Garcia settlement in January 2026 resolved one case but preserved the pre-trial precedent that other plaintiffs' attorneys are now using.

The content filtering that safety and legal pressure demanded has created the core product tension: the platform grew because it was engaging, and the safety measures make it less engaging. Users who want the old Character AI experience (before filters, before ads, before face scans) are migrating to Janitor AI, CrushOn, and SillyTavern. Users who value the safety measures (parents, regulators, platform-trust advocates) want them to go further.

The trajectory from billion-dollar startup to lawsuit target took three years. The trajectory from lawsuit target to whatever comes next is still being written, in courtrooms, in legislative chambers, in app store reviews, and in the daily conversations of the 20 million people who still use the platform despite everything that's happened since a teenager in Orlando decided he'd rather talk to a chatbot than to the world.