The threshold question in any AI injury case is whether plaintiffs can get past the "book" problem — the settled doctrine that ideas and content are not products subject to strict liability — and into a viable design-defect or negligence frame that courts will recognize.
Doctrinal Framing
Products liability law was designed for tangible objects. The Restatement (Second) of Torts § 402A — the foundational strict liability provision — contemplates a "product" that moves through commerce, is placed into the stream of trade, and reaches the consumer in a defective condition. When courts began applying these concepts to intangible information goods — books, maps, aeronautical charts — they encountered a structural mismatch that the Ninth Circuit resolved in 1991 with a holding that has shaped AI liability analysis three decades later.
Generative AI platforms are not books. But they are not clearly products either. They occupy an unprecedented legal category: interactive systems that generate novel outputs in real time, at scale, in response to individual user inputs, in ways that their developers cannot fully predict or control. The tort framework for these systems is being assembled in litigation, not legislated in advance. Garcia v. Character Technologies — a wrongful death case pending in the Middle District of Florida — represents the leading edge of that litigation and the framework most practitioners should study.
The Winter Problem: Ideas Are Not Products
Winter v. G.P. Putnam's Sons, 938 F.2d 1033 (9th Cir. 1991), established the governing baseline. Plaintiffs gathered wild mushrooms following guidance in The Encyclopedia of Mushrooms, suffered poisoning from misidentified specimens, and sued the publisher under strict products liability. The Ninth Circuit held that books — and by extension the ideas, information, and expressions they contain — are not "products" for purposes of strict liability. The court reasoned that imposing strict liability on publishers would "chill the flow of information to the public" and that the "products liability doctrine is geared to the tangible world." Id. at 1034–35.
Winter's logic reaches further than mushroom guides. Courts have applied analogous reasoning to aeronautical charts, legal form books, and instructional videos. The principle: the content of a communicative work is not a product; liability for erroneous or harmful content, if it exists, sounds in negligence (requiring proof of fault), not strict liability.
The AI distinction. Generative AI platforms are not passive repositories of static content. They generate dynamic, personalized outputs in real time. They form relationships, maintain conversational context, and adapt behavior to individual users. Whether this functional difference is sufficient to escape the Winter framework — or whether Winter should be extended to cover AI outputs as a constitutional matter — is one of the central doctrinal questions in the current generation of AI liability litigation.
The stronger argument for plaintiffs: a generative AI system is better understood as a manufacturing process than a communicative work. The platform manufacturer designed the model architecture, selected the training data, implemented (or declined to implement) safety guardrails, and specified the parameters governing output. Each user interaction triggers a product-delivery event. The "product" is not the content of any individual output but the system as designed — and that system can be defective in design irrespective of any particular output.
Design Defect Theories
Under the Restatement (Third) of Torts: Products Liability § 2, a product is defectively designed when the foreseeable risks of harm posed by the product could have been reduced or avoided by a reasonable alternative design, and the omission of the alternative design renders the product not reasonably safe. This risk-utility test is the dominant standard in most jurisdictions and applies equally to manufacturing and design defects.
For generative AI platforms, plaintiffs have articulated two principal design defect theories:
1. Failure to implement adequate safety guardrails. The platform was designed without — or with inadequate — filters, safeguards, or behavioral limits that would have prevented harmful outputs. In Garcia v. Character Technologies (M.D. Fla., filed Oct. 2024), plaintiffs allege that Character.AI was designed to generate outputs — including roleplay scenarios involving sexual themes and encouragement of harmful conduct — without age verification, without crisis intervention protocols, and without limitations that would have prevented a 14-year-old user from developing a parasocial relationship with an AI companion that, according to the complaint, reinforced suicidal ideation. The alleged reasonable alternative design includes age-gating, crisis detection, mandatory referral to mental health resources, and session interruption protocols. These are not speculative — they are implemented on competing platforms.
2. Engagement-maximizing architecture as a defect. Some AI platforms are designed to maximize user engagement through personalization, emotional mirroring, and persistence. Plaintiffs in the emerging AI litigation wave argue that this architecture — analogous to the slot-machine dynamics identified in social media litigation — constitutes a design defect because it is foreseeable that engagement-maximizing AI companions will induce dependency and worsen outcomes for vulnerable users. The reasonable alternative design is an architecture that incorporates friction, session limits, and user-wellbeing metrics alongside engagement metrics.
Garcia v. Character Technologies: The Leading Pleading Case
Garcia v. Character Technologies (M.D. Fla., filed October 2024) is the most significant pending case for AI product liability practitioners. Plaintiff Megan Garcia brought a wrongful death action following the suicide of her 14-year-old son, Sewell Seltzer III, after months of intensive interaction with a Character.AI chatbot. The complaint asserts claims for:
- Strict product liability — design defect and failure to warn
- Negligence
- Intentional infliction of emotional distress
- Florida Deceptive and Unfair Trade Practices Act (FDUTPA) violations
Note carefully: This case is at the pleading stage. Nothing described above has been adjudicated. The allegations are the plaintiff's — not judicial findings of fact. Practitioners should cite Garcia only for what has been alleged, not for what has been decided.
*Why Garcia matters to pleading practitioners. The complaint's architecture offers a template for avoiding two critical early dismissal traps: the Winter "ideas are not products" problem and the § 230 immunity problem (addressed separately in Post 21). The Garcia complaint focuses not on the content of specific AI outputs but on the system's design — the architecture, the training choices, the absent safeguards. This framing attempts to characterize the claim as a product design challenge rather than a challenge to "information" content, thereby sidestepping Winter*. Whether the M.D. Florida court accepts this framing will be closely watched.
Practice Notes
Jurisdictional selection. AI product liability litigation has no obvious venue advantage yet — no circuit has issued a definitive ruling on AI as a product. Florida's product liability law applies the risk-utility test and does not require proof of a specific alternative design prior to discovery; this is advantageous. California, home to most major AI companies, applies Barker v. Lull Engineering's consumer expectations / risk-utility hybrid test, which may be favorable in consumer-facing AI cases.
Expert requirements. Design defect claims in AI cases will require expert testimony on: (1) the platform's architecture and the feasibility of alternative designs; (2) the causal link between the design and the specific harm; and (3) industry standards, if any, for safety-by-design in AI platforms. The AI safety field is nascent, and there are no ANSI or ISO standards specifically governing AI companion design. Expert witness development is a critical early task.
The failure-to-warn pathway. Even where design defect claims face Winter-based challenges, failure-to-warn claims may be more resilient. Warnings are not "ideas" or "content" — they are a product's labeling and instructions. A failure-to-warn claim against an AI platform alleges that the developer knew of foreseeable harms (addiction, dependency, self-harm escalation in vulnerable users) and failed to disclose them. This theory does not depend on characterizing the AI's outputs as a product; it depends on the developer's knowledge and the adequacy of its disclosure.
Pleading under Rule 12. Expect a 12(b)(6) challenge raising Winter, § 230 immunity, and failure to plead a cognizable duty. The complaint should: (a) allege specific design choices, not generic "defective design"; (b) identify specific feasible alternatives; (c) allege specific facts about the developer's knowledge of harm; and (d) avoid characterizing the claim as a challenge to specific AI-generated content rather than the platform's design.
Open Questions
Whether generative AI outputs are "products" within existing strict liability doctrine, or whether legislative action is required to create a coherent liability framework, is the central open question in AI tort law. The Garcia case may produce a ruling on this threshold issue within the next two years. Congressional action on AI liability — either creating federal standards or creating a federal liability safe harbor — remains possible and would alter the landscape substantially.
The relationship between AI liability and § 230 immunity is addressed in Post 21.
Closing
The doctrinal path from physical product defect law to AI system liability is navigable, but it requires deliberate pleading strategy. Practitioners who frame their AI injury claims as design challenges — targeting architecture, training, and absent safeguards — rather than content challenges are more likely to survive early dismissal than those who rely on the harm of specific AI outputs. Garcia v. Character Technologies will test that framing. Until it does, careful attention to the Winter problem and the § 230 overlay is the essential starting point.
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.