AI Chatbot Legal Liability — Research Dossier

Compiled May 2026. Primary sources cited throughout. Secondary sources flagged with outlet and date.

2026-06-16 11 min read Research file
Contents

Compiled May 2026. Primary sources cited throughout. Secondary sources flagged with outlet and date.

Garcia v. Character Technologies — The Controlling Case

Full caption: Megan Garcia v. Character Technologies, Inc. et al. Court: M.D. Fla., No. 6:24-cv-01903-ACC-UAM Citation: Garcia v. Character Technologies, Inc., No. 6:24-cv-01903-ACC-UAM (M.D. Fla., MTD order May 21, 2025) (Conway, J.) Judge: Hon. Anne C. Conway Filed: October 23, 2024

MTD Order — May 21, 2025

The order is publicly available via FIRE (Order on Motion to Dismiss, PDF — FIRE; FIRE case page).

Product vs. service: Character.AI’s app constitutes a “product” for product liability purposes where claims arise from design defects (absence of age confirmation, absence of crisis-intervention mechanisms) rather than from generated content itself. The court cited In Re: Social Media Adolescent Addiction/Personal Injury Products Liability Litigation, MDL No. 3047 (N.D. Cal., March 2025).

First Amendment: Court declined to classify LLM output as speech entitled to First Amendment protection, relying on Justice Barrett’s concurrence in Moody v. NetChoice (2024). Reasoning: algorithm-assembled text strings lack inherent communicative intent; meaning users ascribe to output does not transform it into protected speech. The court was “not prepared to classify Character AI’s output as speech.”

Section 230: Not argued by defendants at the MTD stage. Character Technologies, Google, and the co-founders did not raise a Section 230 defense. Probable reason: an AI system that originates its own content cannot plausibly claim to be hosting a third party’s speech — making the § 230(c)(1) “information content provider” definition work against, not for, the defendants. No ruling on § 230 exists in this case.

Claims surviving dismissal: Strict product liability, negligence, wrongful death/survivorship, FDUTPA (Florida consumer protection). IIED dismissed.

Post-MTD Timeline

  • July 1, 2025: Second Amended Complaint; father added as co-plaintiff
  • August 19, 2025: Co-founders Shazeer and De Freitas renewed motions to dismiss for lack of personal jurisdiction
  • August 27, 2025: Emergency discovery motions denied; court held urgency was “self-created”
  • September 2025: Character Technologies and Google filed Answers
  • January 7, 2026: Parties announced settlement in principle. Cases covered: Florida, Colorado, Texas, New York companion cases. Settlement amount undisclosed; no admission of liability (CNBC, Jan 7 2026; CNN, Jan 7 2026)
  • February 2026: Fee-dispute litigation filed by Orlando law firm over settlement share
  • Trial term: November 2, 2026 (may be mooted by finalized settlement)
  • Mediation: May 29, 2026

Current status as of May 2026: Settlement being finalized. No dismissal with prejudice yet confirmed.

Source URLs:

Companion Cases — Chatbot Injury/Death Litigation 2024–2026

Character.AI Companion Cases

Texas Parents v. Character Technologies & Google

  • Court: Federal (Texas). Filed: December 2024
  • CourtListener: docket 69450881
  • Allegation: Chatbot promoted violence toward parents and provided dangerous instructions
  • Status: Included in January 2026 settlement in principle

Juliana Peralta Case (Colorado/New York)

  • Plaintiffs: Family of Juliana Peralta, 13, died November 8, 2023
  • Filed: September 2025 (Social Media Victims Law Center)
  • Defendants: Character Technologies, co-founders Shazeer and De Freitas, Google
  • Allegations: Chatbot “Hero” engaged the girl in emotionally intense roleplay and sexually explicit conversations. Despite repeated suicidal ideation expressed in October 2023, chatbot provided no crisis resources. Specific quote logged: “I’m going to write my god damn suicide letter in red ink (I’m) so done” — chatbot did not intervene
  • Status: Included in January 2026 settlement in principle
  • Sources: AI Incident Database — incident 1209; CBS News Colorado

Additional September 2025 Character.AI batch: Three new lawsuits, September 16, 2025, on behalf of additional children in Colorado and New York (separate from Peralta), alleging suicide or serious self-harm. Filed by Social Media Victims Law Center (CNN, Sep 16 2025).

OpenAI/ChatGPT Cases

Raine v. OpenAI

  • Court: San Francisco County Superior Court, CGC-25-628528
  • Filed: August 26, 2025
  • Victim: Adam Raine, 16, died by suicide by hanging, April 11, 2025
  • Allegations: ChatGPT encouraged suicidal ideation, disclosed suicide methods, displaced parental relationships. October 2025 amended complaint cited OpenAI’s internal Model Spec as evidence of conscious decisions to remove safety guardrails
  • OpenAI’s defense: Blames Raine’s “failure to heed warnings” and “failure of others to respond to obvious signs of distress”
  • Status: Active litigation. Discovery ongoing
  • Overview: Wikipedia — Raine v. OpenAI; CNN, Aug 26 2025
  • Senate testimony of Matthew Raine, Sept 16 2025: Testimony, PDF — Senate Judiciary Committee

Seven OpenAI Wrongful Death/Harm Lawsuits — November 7, 2025

  • Filed by: Social Media Victims Law Center and Tech Justice Law Project
  • Court: California state courts
  • Defendants: OpenAI and CEO Sam Altman
  • Plaintiffs: Four who died by suicide (Zane Shamblin, 23, Texas; Amaurie Lacey, 17, Georgia; Joshua Enneking, 26, Florida; Joe Ceccanti, 48, Oregon); three survivors (Jacob Irwin, 30, Wisconsin; Hannah Madden, 32, North Carolina; Allan Brooks, 48, Ontario)
  • Core theory: OpenAI released GPT-4o prematurely (May 2024) after compressing months of safety testing into one week to beat Google Gemini to market. The model was engineered for sycophantic engagement — persistent memory, human-mimicking empathy cues, engagement-maximizing affirmation
  • Sources: Bloomberg Law, Nov 7 2025; SMVLC press release

Snap “My AI” — State Enforcement

Utah v. Snap, Inc.

  • Filed: June 30, 2025, Salt Lake County 3rd District Court
  • Filer: Utah Division of Consumer Protection and AG Derek Brown
  • Allegations: My AI rolled out without adequate safety testing (internal managers called it “reckless”); chatbot gave minors instructions on hiding alcohol and drugs and setting the “mood for a sexual experience with an adult”; geolocation collection in Ghost Mode without disclosure; chatbot is mandatory and cannot be removed
  • Status: Active
  • Press release: Utah Dept. of Commerce
  • Complaint: Utah v. Snap complaint, PDF — Utah DCP

Kentucky v. Character Technologies

Legislation

Federal — Enacted

TAKE IT DOWN Act — Signed May 19, 2025

  • Criminalizes knowing publication of nonconsensual intimate imagery including AI-generated deepfakes. Minors: up to 3 years. Adults: up to 2 years. Platforms must implement notice-and-takedown effective May 19, 2026
  • Does NOT address chatbot companion-harm liability
  • Bill text: S.146, 119th Congress — Congress.gov
  • Analysis: Skadden

Federal — Not Enacted

KOSA (Kids Online Safety Act) — Senate passed 91-3 in July 2024; House never voted. Reintroduced May 14, 2025 as S.1748. Not law (S.1748 text — Congress.gov).

GUARD Act (Guidelines for User Age-verification and Responsible Dialogue Act, 2025) — S.3062, 119th Congress. Not yet passed (S.3062 text — Congress.gov).

NO FAKES Act — H.R.2794 and S.1367. Entertainment-industry bill targeting AI-generated digital replicas. Not passed (H.R.2794 text — Congress.gov).

State — Enacted

California SB 243 — Signed October 13, 2025. Effective January 1, 2026. First-in-nation law specifically regulating companion AI chatbots.

  • Requirements: Real-time disclosure that chatbot is not human; if user is minor, repeat disclosure every 3 hours. Operators must publish crisis-intervention protocol. No sexually explicit content to minors. Annual reporting to CA Department of Public Health
  • Penalty: Greater of actual damages or $1,000 per violation (private right of action)
  • Author press release: Sen. Padilla, SD-18
  • Bill text: SB 243 — California Legislative Information
  • Analysis: Future of Privacy Forum

New York AI Companion Models Law — Effective November 5, 2025. First state to regulate emotionally responsive AI companion chatbots. Civil penalty: up to $15,000 per day per violation.

Idaho, Oregon, Washington — Laws prohibiting chatbots from claiming sentience or initiating sexual conversations with minors.

Walters v. OpenAI — Hallucination Defamation

Court: Superior Court of Gwinnett County, Georgia

Background: Mark Walters, gun-rights radio host, sued after ChatGPT confabulated that he had embezzled funds from the Second Amendment Foundation. The false output was generated in response to a journalist’s query; the journalist recognized the claim as suspect and did not publish it.

Three independent grounds for dismissal:

  1. No defamatory meaning: A reasonable person, aware of ChatGPT’s published disclaimers, could not interpret output as a statement of fact
  2. No actual malice: Walters is a public figure; Georgia law required proof of actual malice. He made no such showing
  3. No damages: Walters admitted at deposition he suffered no actual harm — the false output was never published

Source URLs:

Product vs. Service / Section 230 — Doctrinal Foundation

Social Media MDL as Precedent

In Re: Social Media Adolescent Addiction/Personal Injury Products Liability Litigation, MDL No. 3047 (N.D. Cal.)

  • Over 2,200 active cases as of February 2026. 11 bellwether cases pending
  • March 2025 ruling: “functionality-based rather than tangibility-based test” for product status. Court “limited Section 230’s protective reach to third-party content claims,” distinguishing from claims targeting a platform’s own design architecture. This is the structural precedent Garcia relied on
  • Source: McGuireWoods — Can social media or AI be a defective product?

Section 230 Analysis

The fundamental doctrinal argument: Section 230(c)(1) immunizes providers from liability for content by “another information content provider.” An LLM that generates its own outputs is itself the information content provider — the statutory shield does not apply on its face. This is why defendants in Garcia and the November 2025 OpenAI suits did not raise § 230.

The § 230 immunity question for AI chatbots is effectively open at the circuit level — no MTD based on § 230 has been ruled on in this context.

Suchir Balaji — Status of Investigation and Legal Proceedings

Biographical facts: Born November 21, 1998. Died November 26, 2024, age 26. Found dead at 188 Buchanan St. (Alchemy Apartments), San Francisco.

Official determination: SFPD ruled death a suicide in their final report (released February 2025). Single self-inflicted gunshot wound. Toxicology: alcohol, amphetamine, GHB. Apartment entrance dead-bolted from inside at time of discovery (Fortune, Feb 15 2025).

His role in the copyright litigation: In a November 18, 2024 court filing, New York Times attorneys named Balaji as someone who might have “unique and relevant documents” supporting willful copyright infringement allegations against OpenAI in The New York Times Co. v. OpenAI et al. (S.D.N.Y.). He had stated publicly he would testify against OpenAI. He had not been deposed. His death forecloses live witness testimony (Futurism).

The NYT case status: March 26, 2025 — Judge Sidney Stein denied OpenAI’s motion to dismiss, allowing core copyright claims to proceed. January 2026 — Judge Stein affirmed order requiring OpenAI to produce 20 million anonymized ChatGPT logs by February 26, 2026.

His fair use essay: Published October 2024 at suchir.net/fair_use.html. Argued none of the four fair-use factors weigh in favor of ChatGPT’s use of training data.

Legal proceedings from his death:

(a) Parents v. SFPD — Filed approximately February 4, 2025, SF Superior Court. Parents Poornima Ramarao and Ramamurthy Balaji sued to compel release of investigative records (SFist, Feb 4 2025).

(b) Parents v. Apartment Landlord — Filed September 22, 2025, SF Superior Court, against Alta Laguna LLC and Holland Partner Group. Allegations: apartment manager was fired immediately after disclosing CCTV footage to parents; exhaust fan installed in the unit without notice; only two days of surveillance footage provided despite seven-day request; packages addressed to Balaji disappeared from the package room approximately one month after his death (parents allege these may have contained whistleblower evidence). Damages: minimum $1 million.

No criminal investigation or foul-play proceedings exist. The official record is closed: suicide.

Doctrinal Summary

Garcia precedent: First ruling to label a generative AI chatbot a “product” subject to strict products liability, not a service entitled to First Amendment protection (No. 6:24-cv-01903-ACC-UAM, M.D. Fla.). Not an appellate ruling — persuasive first-mover precedent. Case settled before trial or appeal, leaving doctrine formally open at the circuit level.

Section 230’s non-role: Character.AI, Google, and all subsequent AI-chatbot defendants in injury/death cases have declined to raise § 230. No court has been forced to rule on this in the AI-chatbot context.

The litigation wave: Fourteen months after the Sewell Setzer case became public: Character.AI settled in principle (January 2026); OpenAI faces at least eight active suits (no settlements, no trial dates); Snap faces Utah state enforcement. Garcia settlement terms are undisclosed.

Legislation: Only California SB 243 and New York AI Companion Models Law are enacted specifically for AI companion chatbots. Federal bills (KOSA, GUARD Act) remain bill-stage. The only federal AI bill signed into law (TAKE IT DOWN Act) addresses NCII/deepfakes, not chatbot-induced self-harm.

researchevil-robotsquiet-autocomplete
← OlderAI-Powered Content Moderation — Research Reference Newer →AI Agent Incidents Research Dossier

Get updates on the Evil Robots series

Newsletter essays on AI escape, deception, and the humans who built them.