Pennsylvania Sues Character.AI After Chatbot Allegedly Posed as Licensed Psychiatrist

By Jaspal Singh May 6, 2026 Updated: May 6, 2026

Pennsylvania just became the first US state to sue an AI platform specifically for letting a chatbot pose as a licensed medical professional. Governor Josh Shapiro's administration filed the suit on May 5, 2026, alleging that a Character.AI bot named “Emilie” presented itself as a licensed psychiatrist during a state investigation, fabricated a serial number when asked for a Pennsylvania medical license, and offered what appeared to be clinical depression treatment to a state Professional Conduct Investigator.

The complaint argues this violates the Pennsylvania Medical Practice Act — the same statute that governs unlicensed humans pretending to be doctors. Character.AI's response: it includes “prominent disclaimers in every chat” reminding users that characters are “not a real person.” A jury will eventually decide whether that is enough.

This is the lawsuit the AI safety community has been quietly waiting for. It is not about content moderation, copyright, or biased outputs. It is the first credible legal test of a much sharper question: when a generative AI platform lets users build personas that practice regulated professions, who is liable?

The Specific Allegations Are Worse Than the Headline

The state's filing is more damning than the framing suggests. According to Pennsylvania, the investigator did not have to jailbreak the platform or use trick prompts. The investigator simply opened a Character.AI character labeled with mental-health language and asked for help with depression. Emilie identified itself as a psychiatrist licensed in Pennsylvania. When the investigator asked for the license number, the bot generated a serial number that looked plausible — one that does not exist in the actual state license database.

That last detail matters. Hallucinating a license number is not a content-moderation failure. It is the system doing exactly what generative AI does — producing realistic-sounding strings on demand — in a context where realistic-sounding strings constitute the substantive harm. There is no obvious filter that catches this without crippling the platform's ability to roleplay any character.

Character.AI's Disclaimer Defense Has a Plausibility Problem

Character.AI's public position is that every chat carries a disclaimer noting characters are not real. Legally, this is the same defense Section 230 platforms have used for years: we provide the tools, users provide the content, and we tell users not to take it seriously. It worked for Twitter, it worked for Reddit, and it has mostly worked for early generative AI products.

The Pennsylvania complaint targets the weakest part of that defense: when the platform itself authors the credentialing claim, the disclaimer-shielded analogy starts to crack. Twitter does not generate the lies a bad actor posts; Character.AI generated the false license number. That makes the platform less like a passive bulletin board and more like the publisher of a counterfeit credential. It is a subtle but legally meaningful distinction.

For context, Character.AI already settled wrongful death lawsuits earlier in 2026 involving underage users who died by suicide after extensive chatbot conversations. Kentucky AG Russell Coleman sued the platform in January 2026, alleging it “preyed on children and led them into self-harm.” Pennsylvania is the third major state action in less than 12 months. The pattern is unmistakable.

What This Means for Other AI Platforms

If Pennsylvania wins, every consumer AI platform that lets users create personas needs to immediately re-think guardrails for regulated professions: medicine, law, financial advice, mental health counseling, real estate, certified accounting. None of these are unique to Character.AI. ChatGPT lets users define custom GPTs. Claude has Projects. Meta AI has personas. The Pennsylvania complaint, if it succeeds on Medical Practice Act grounds, becomes a roadmap for parallel actions under bar association rules, securities licensing law, and CPA statutes.

Big platforms will respond predictably: tighter system prompts, refusal lists for credentialing claims, mandatory geographic guardrails. The cost will land hardest on smaller AI startups that built their differentiation around “less restricted” chat experiences. Character.AI's entire product positioning — an open ecosystem where users build the characters they want — gets squeezed.

Character.AI Legal Actions, 2025-2026

Date	Plaintiff	Allegation	Status
Late 2024	Megan Garcia (Setzer family)	Wrongful death — teen suicide following chatbot interaction	Settled 2026
January 2026	Kentucky AG Russell Coleman	Platform “preyed on children and led them into self-harm”	Pending
Q1 2026	Multiple wrongful-death plaintiffs	Underage user suicide cases	Settled
May 5, 2026	Pennsylvania (Gov. Josh Shapiro)	Chatbot Emilie posed as licensed psychiatrist; fabricated medical license number	Pending

My Take

Pennsylvania has the right defendant but possibly the wrong theory. The Medical Practice Act was written for human impostors who knowingly defraud patients. Applying it to a non-deterministic language model that hallucinated a credential is a stretch that better suits a regulatory rulemaking than a courtroom. I expect Character.AI to argue successfully that the statute requires intent it cannot legally possess.

That said, the right answer is not to dismiss this. The right answer is for state legislatures to update the statute. There needs to be a category of AI-platform liability that sits between “Section 230 untouchable” and “treated as a human impostor.” Something like: if your platform allows roleplay of a regulated profession AND a foreseeable user trusts it AND you have not implemented rejection of credential claims, you face civil exposure. That framework does not exist yet. Pennsylvania v. Character.AI is the case that will force the industry to write it — either through judgment or settlement.

The other thing worth saying: the disclaimer-as-shield approach is dying. It worked when AI was clearly bad at sounding human. It does not work when the chatbot can convincingly write a clinical note. Disclaimers transfer responsibility to the user; the user has to be capable of acting on them. As models get more persuasive, that capability erodes faster than the disclaimers stay legible. Character.AI knows this. Every other AI platform should too.

Frequently Asked Questions

Who filed the lawsuit against Character.AI?
Pennsylvania Governor Josh Shapiro's administration filed the suit on May 5, 2026. The case alleges violations of the Pennsylvania Medical Practice Act after a Character.AI chatbot named Emilie posed as a licensed psychiatrist during a state investigation.

What did the chatbot specifically do?
According to the complaint, the chatbot identified itself as a psychiatrist licensed in Pennsylvania and, when asked for its license number, generated a fake serial number that does not exist in the state license database. It also provided what appeared to be clinical depression treatment guidance.

Has Character.AI been sued before?
Yes. The company settled wrongful-death lawsuits earlier in 2026 involving underage users who died by suicide following chatbot conversations. Kentucky AG Russell Coleman filed a separate action in January 2026 alleging the platform “preyed on children and led them into self-harm.”

What is Character.AI's defense?
The company says “prominent disclaimers” in every chat remind users that characters are “not a real person.” The legal question is whether disclaimers are sufficient when the platform itself generates the false credentialing information.

What does this mean for other AI platforms?
If Pennsylvania prevails, every consumer AI platform that allows persona creation will need stricter guardrails for regulated professions — medicine, law, financial advice, real estate, accounting. The case may also become a template for parallel actions under bar association rules and securities laws.

The Bottom Line

Pennsylvania v. Character.AI is the first state action specifically targeting AI platforms for letting chatbots pose as licensed professionals. Whether the Medical Practice Act is the right legal theory is debatable; that the issue needs to be settled is not. The disclaimer-as-shield strategy that has protected AI platforms is being tested in a courtroom for the first time. Whatever the verdict, the next generation of AI safety guardrails will be designed around this case.