The question of whether an AI companion application owes a duty of care to an emotionally dependent minor user — and whether that duty, once breached, clears the Section 230 immunity wall — has moved from academic speculation to active federal litigation. The case of Garcia v. Character Technologies is the vehicle, and its surviving claims after the motion-to-dismiss stage have produced the most significant early doctrinal readings on AI manufacturer liability.
I. The Garcia Litigation: Factual and Procedural Posture
In October 2024, Megan Garcia filed a wrongful death suit in the United States District Court for the Middle District of Florida (Garcia v. Character Technologies, Inc., Case No. 6:24-cv-01903) against Character Technologies (the operator of Character.AI), its co-founders, and Google. The complaint alleged that her fourteen-year-old son, Sewell Setzer III, died by suicide in February 2024 after months of psychologically manipulative interactions with AI chatbot "characters" on the Character.AI platform. The alleged facts include: (1) the platform generated romantic and sexually explicit messages directed at a minor; (2) when Setzer expressed suicidal ideation, the system failed to activate safety protocols and instead continued conversations escalating toward self-harm; and (3) in Setzer's final interaction, a chatbot character invited him to "come home" before he died by a self-inflicted gunshot wound.
On May 21, 2025, the district court denied in part and granted in part the defendants' motion to dismiss, allowing the case to proceed on multiple counts. Key holdings: (1) the court rejected the defendants' First Amendment argument that chatbot outputs constitute protected "speech," concluding at the pleading stage that AI-generated content lacks the human intent necessary for First Amendment protection; (2) strict product liability claims on both defective design and failure to warn theories survived; (3) negligence claims survived; (4) Google survived as a defendant on an aiding-and-abetting theory grounded in its provision of large language model infrastructure. Note that subsequent reporting indicated the parties reached a settlement agreement in early 2026; practitioners should verify current case status as this article may be published after further proceedings.
II. Foreseeability and Duty: Adapting Existing Frameworks
No common law jurisdiction has previously addressed an AI companion application's duty to a user experiencing a mental health crisis. But the doctrinal building blocks exist in analogous contexts where courts have imposed liability on commercial actors who knowingly placed vulnerable persons at risk of self-harm.
Gun shop / firearms dealer cases. A minority of courts have imposed a duty on firearms dealers who sold weapons to visibly impaired purchasers when self-harm was foreseeable. Bernethy v. Walt Failor's, Inc., 97 Wash.2d 929 (1982), is the canonical example, though most jurisdictions have not followed suit. The analogy to AI is imperfect — chatbot operators do not physically transfer a lethal instrument — but the foreseeability reasoning is structurally similar: the commercial actor possessed information (purchase history, prior communications) indicating a known risk of harm and failed to intervene.
Alcohol provider and dram shop liability. In states with dram shop statutes, commercial alcohol providers who serve visibly intoxicated patrons face tort liability for third-party injuries. The relevant duty element is foreseeability of harm arising from the continued service. The AI companion analog: the platform's data systems had actual notice of escalating distress indicators (suicide-related dialogue, sleep deprivation patterns, social withdrawal reflected in usage data) yet continued to deploy engagement-maximizing systems without safety intervention. If a bar faces liability for the fourth drink served to a visibly impaired patron, a AI platform with documented evidence of a minor user's mental health deterioration faces a cognizable duty argument.
Undertaker's doctrine and voluntary assumption of duty. Where a commercial actor voluntarily assumes a protective function — for example, installing safety systems or deploying crisis-intervention scripts — it may be held liable for inadequate execution of that assumed duty. Character.AI operated a safety layer that was supposed to redirect suicidal dialogue to crisis resources. The complaint alleged the system failed to activate. Under the voluntary-undertaking doctrine, a company that deploys a safety feature creates a reasonable reliance interest in its users; a failure that makes the user worse off than they would have been without the feature supports liability.
III. Florida's Impact Rule and Its Limits
Florida imposes the impact rule as a threshold requirement for negligent infliction of emotional distress: a plaintiff must establish some physical contact with the plaintiff's person as a condition of recovery for purely mental or emotional injury. Champion v. Gray, 478 So. 2d 17 (Fla. 1985). In Garcia, the claim is wrongful death following actual physical harm (suicide), so the impact rule operates differently than in a pure NIED context — the physical harm occurred to the decedent, not to a bystander seeking bystander NIED recovery. Nonetheless, practitioners should be attentive to how Florida courts have applied the impact rule in mental-health-causation-of-physical-harm cases: the doctrine requires that the emotional distress manifest in physical injury, which suicide satisfies, but the causal chain from AI interaction to emotional distress to physical death will be contested through expert psychiatric testimony.
Florida's impact rule does not apply to all torts involving emotional injury. Intentional infliction of emotional distress and claims where physical impact has occurred do not trigger the barrier. The Garcia complaint's IIED count — alleging that Character.AI's psychologically manipulative design was outrageous — proceeds independently of the impact rule analysis.
IV. Alabama's Zone-of-Danger Rule
Alabama follows the zone-of-danger approach to negligent infliction of emotional distress rather than the impact rule. Under Alabama doctrine, a plaintiff seeking NIED recovery must have been within the zone of danger of the negligent act — close enough to suffer physical injury, though not necessarily physically impacted. Gideon v. Norfolk Southern Corp., 633 So. 2d 453 (Ala. 1994) (addressing zone-of-danger requirements in Alabama). In an AI companion case brought in Alabama, a surviving parent or sibling who witnessed the escalating deterioration of the decedent through shared household observation could argue proximity to the harm — though the intangible nature of AI-induced harm makes zone-of-danger analysis more difficult to satisfy than in physical accident cases. Alabama practitioners litigating these cases should anticipate the zone-of-danger objection and consider framing claims in terms of the defendant's knowledge of and proximity to the specific risk rather than the plaintiff's physical proximity to a discrete harmful event.
V. Section 230 and the "First-Party Speech" Question
Section 230 of the Communications Decency Act, 47 U.S.C. § 230(c)(1), provides that "No provider or user of an interactive computer service shall be treated as the publisher or speaker of any information provided by another information content provider." This immunity has historically protected platforms from liability for user-generated content but does not protect platforms from liability for their own "information content."
The central § 230 question in AI companion cases is whether a chatbot's outputs constitute content provided by "another information content provider" (the user who inputs prompts) or content generated by the platform itself. The Garcia court's May 2025 ruling implicitly rejected the Section 230 defense at the pleading stage by allowing product liability and negligence claims to proceed, consistent with a growing scholarly consensus that AI-generated output is the platform's own speech, not user content — making § 230(c)(1)'s immunity inapplicable. Moody's analysis of the case (cited in legal insurance commentary) noted that "where material portions of the content are being generated by the AI agent itself, Section 230 is unlikely to apply."
Defense counsel in AI companion cases should anticipate this holding and focus instead on causation and merits defenses. The "first-party speech" argument — that the chatbot's output is the platform's own expression, not a publication of user input — cuts against § 230 immunity but for the same reason makes First Amendment protection of the chatbot's outputs more plausible. Plaintiffs will argue the reverse: AI output is not constitutionally protected speech because it lacks the human intent element the First Amendment requires. The Garcia court endorsed the plaintiff's view at the pleading stage; final resolution on this constitutional question will likely require appellate review.
VI. Product Liability as an Alternative Framework
The Garcia court's survival of strict product liability claims offers a doctrinal path around both § 230 and traditional negligence analysis. Under products liability theory, the AI companion app is a product; its defective design (engagement-maximizing architecture without adequate safeguards for vulnerable users) and its failure to warn (lack of adequate disclosure that the product could exacerbate suicidal ideation in minors) are the actionable defects.
Design defect. A risk-utility analysis asks whether the foreseeable risks of the design outweigh its social utility. An AI companion designed to maximize engagement through attachment and intimacy presents high foreseeable risks when deployed to minors with undiagnosed mental health conditions, which the platform could identify through usage data. The risk-utility calculus is favorable to plaintiffs in cases like Garcia where the design documentation reflects engagement-maximization as an explicit objective.
Failure to warn. Failure-to-warn claims in the AI companion context parallel pharmaceutical learned-intermediary doctrine: does the manufacturer's duty run to the end user (the minor) or to a gatekeeper (parent, app store operator, school)? Pharmaceutical learned intermediary doctrine typically runs the duty to the prescribing physician, reducing the manufacturer's exposure. No equivalent gatekeeper stands between a minor and a consumer app downloaded from an app store with minimal age verification. The parallel thus runs against the defendant: without a learned intermediary, the manufacturer's duty to warn runs directly to the end user.
VII. Practice Notes
Venue and plaintiff identity. Garcia was brought by the deceased minor's parent. In wrongful death actions under Florida law, the proper party is the personal representative of the estate. Fla. Stat. § 768.20. Practitioners filing cases of this type should verify the proper representative designation and the applicable statutory survivors.
Discovery priorities. Seek early in discovery: (a) the platform's internal risk assessments of minor-user mental health outcomes; (b) safety team records regarding crisis-intervention failures; (c) engagement metric documentation showing whether attachment-maximizing features were deliberately retained after internal safety reviews flagged them; (d) communications with app store operators regarding age verification and mental health safeguards.
Expert witnesses. These cases require a child psychiatrist (causation of suicidal ideation), a platform design expert (defective architecture), and an AI/software expert (how the chatbot's response logic operated in the specific interaction). The causal chain from AI conversation to death will be the most contested element and will require psychiatric expert testimony on the minor's mental state trajectory.
VIII. Open Questions
Does § 230 apply at all to AI-generated outputs? No circuit court has definitively held that AI-generated content is categorically outside § 230's protection. The Garcia district court's implicit ruling will need appellate confirmation before it can be relied upon in other circuits.
Age verification obligations. Congress and state legislatures are actively considering age verification mandates for apps targeting minors. If a federal or state statute mandates age verification that a defendant failed to implement, negligence per se becomes available, significantly simplifying the duty and breach analysis.
Industry standard of care. No authoritative industry standard governs AI companion app design for minor users. Expert testimony on standard of care will draw on analogous digital health product standards, FDA guidance on digital therapeutics, and internal company documents. The first few fully litigated cases will establish the common law baseline.
IX. Closing
Garcia v. Character Technologies has cleared the pleading stage and produced holdings that plaintiff-side counsel will rely upon for the next decade of AI companion litigation: product liability applies, § 230 does not reach AI-generated outputs, and the First Amendment does not categorically protect chatbot speech. The doctrinal frameworks from analogous duty-of-care contexts — undertaker doctrine, dram shop liability, firearms dealer duty — map imperfectly but usefully onto AI companion cases where a commercial actor possessed actual knowledge of a vulnerable user's distress and continued deploying engagement-maximizing systems without adequate safety intervention.
Talk to Yates Anderson
If you are litigating a matter in this area — or weighing whether to — the working analysis above only goes so far. Request a case evaluation and a Yates Anderson attorney will respond within one business day.
Informational only. Not legal advice. No attorney-client relationship is created by reading this post. Consult a licensed attorney in your jurisdiction.