AI, Software, and the Boundaries of Strict Liability
The threshold question in any products liability case is deceptively simple: is the thing at issue a "product"? For defense counsel representing AI developers, software publishers, and digital platform operators, this question has become the most consequential gateway in modern tort law.
As the Drug & Device Law Blog observed in launching its landmark fifty-state survey, the "somewhat esoteric issue of whether intangible items — chiefly computer software, website algorithms, and other electronic information — is treated as a 'product' for purposes of imposing strict liability on their creators" has moved from academic curiosity to live litigation. 1
The stakes are enormous. Strict products liability under Restatement (Second) of Torts § 402A imposes liability on anyone who "sells any product in a defective condition unreasonably dangerous to the user or consumer or to his property," regardless of whether "the seller has exercised all possible care in the preparation and sale of his product." 2 If AI output is a "product," its creators face no-fault liability for harms their systems cause. If it is not — if it constitutes information, expression, or a service — then plaintiffs must prove negligence, a substantially higher burden.
This distinction was thrown into sharp relief by Judge Anne Conway's May 2025 ruling in Garcia v. Character Technologies, where the court allowed product liability claims against an AI chatbot developer to proceed, ruling that "Character.AI is a product for the purposes of product liability claims, and not a service." 3 The case arose from the suicide of fourteen-year-old Sewell Setzer III after extensive interaction with a Character.AI chatbot. 4 The American Enterprise Institute's Clay Calvert responded that the ruling "perhaps surprisingly and certainly unfortunately" declined to hold that chatbot output constitutes speech protected by the First Amendment. 5
Yet the Garcia order stands against decades of precedent holding that ideas, expression, and information are not products. As Asaf Lubin documented in his comprehensive 2025 Indiana Law Journal essay, "despite software's ubiquity in modern life, its classification within product liability law remains unsettled," with "[j]udicial decisions further complicat[ing] this landscape, with courts diverging on whether software qualifies as a product." 6 His fifty-state survey revealed "widespread doctrinal inconsistencies within the common law"ent.cgi?article=11587&context=ilj" target="_blank" class="quote-link" title="Verify quote in source (PDF)">"widespread doctrinal inconsistencies within the common law" on this foundational question. 7
For defense counsel, this doctrinal unsettlement is both risk and opportunity. The product line — the boundary between what strict liability reaches and what it does not — is moving. As one commentator noted, "[f]or decades, the prevailing view has been that software is not a 'product.' With the proliferation of mobile apps, social media, and AI-enabled tools, however, that consensus is shifting." 8 Courts are "increasingly willing to entertain product liability claims against software-based systems — either in whole or in part — under a few different theories." and "[a] trio of approaches to software as a product are emerging." 9
This module traces the doctrinal architecture of the non-product defense across five eras of development, from the Restatement foundation through the aeronautical chart cases to today's AI litigation, equipping defense counsel with the historical depth, jurisdictional specificity, and strategic frameworks needed to argue that their client's AI system, software, or digital output is not a "product" subject to strict liability.
The modern edifice of strict products liability rests on § 402A of the Restatement (Second) of Torts, adopted by the American Law Institute in 1965. Henderson and Twerski, the reporters who would later propose its revision, observed that "[o]nly rarely do provisions of the American Law Institute's Restatements of the Law rise to the dignity of holy writ" — yet § 402A achieved precisely that status. 10
Section 402A provides that "[o]ne who sells any product in a defective condition unreasonably dangerous to the user or consumer or to his property is subject to liability for physical harm thereby caused to the ultimate user or consumer." 11 Crucially, this liability attaches "although the seller has exercised all possible care in the preparation and sale of his product" and though "the user or consumer has not bought the product from or entered into any contractual relation with the seller." 12
What § 402A did not do was define "product." As Madden noted, the section's "erudite accompanying comments" guided courts through three decades of development, but the text itself assumed a tangible object world. 13 The first decision to hold manufacturers strictly liable in tort for defective products was Greenman v. Yuba Power Products, Inc., and "scores of courts have followed the holding enumerated in Greenman and 'codified' by Restatement (Second) of Torts § 402A." 14
The Third Restatement, completed in May 1997, addressed what its predecessors had assumed. 15 It "goes beyond the law as the law otherwise would stand" and constitutes "an almost total overhaul" of the Second Restatement's products liability provisions. 16
Section 19 of the Third Restatement defines a product as "tangible personal property distributed commercially for use or consumption." 17 The Drug & Device Law Blog emphasized this language: "Other items, such as real property and electricity, are products when the context of their distribution and use is sufficiently analogous to the distribution and use of tangible personal property." 18 But for intangible information, the Restatement struck a different note: courts "'have, appropriately refused to impose strict product liability' where a plaintiff complains of 'the information, not [of] the tangible medium.'" 19
Black's Law Dictionary reinforces this framework, defining "product" as "something that is distributed commercially for use or consumption and that is usu[ally] (1) tangible personal property, (2) the result of fabrication or processing, and (3) an item that has passed through a chain of commercial distribution before ultimate use." 20
The Restatement's notes further explain the rationale: "fear of tort liability could chill the expression of intangible ideas, should strict liability attach to information contained in books, newspapers, movies and other publications or forms of public media." 21 This concern — that extending products liability to information would collide with First Amendment values — became the doctrinal cornerstone of the non-product defense.
The sales chapter of the Uniform Commercial Code defines "goods" — not "products" — as "all things (especially manufactured goods) which are movable at the time of" identification to a contract, with the UCC chapter on leases including the "movable" language but excluding "general intangibles." 22 This parallel body of commercial law reinforces the tangibility premise: the transactional frameworks governing sales of goods were designed for physical objects, not information flows.
The foundational case for the non-product defense is Winter v. G.P. Putnam's Sons, a Ninth Circuit decision that has anchored the information-products boundary for over three decades.
The plaintiffs were "mushroom enthusiasts who became severely ill from picking and eating mushrooms after relying on information in The Encyclopedia of Mushrooms, a book published by the defendant." 23 The district court granted summary judgment for the publisher, and the Ninth Circuit affirmed. 24
The court's reasoning established the doctrinal framework that defense counsel continue to invoke. Writing for the panel, Judge Sneed drew a fundamental distinction between the physical and informational dimensions of a book: "A book containing Shakespeare's sonnets consists of two parts, the material and print therein, and the ideas and expression thereof." 25 The court found that "[p]roducts liability law is geared to the tangible world" and that "[t]he language of products liability law reflects its focus on tangible items." 26
Critically, the court articulated a purpose-based rationale for excluding information from strict liability: "The purposes served by products liability law also are focused on the tangible world and do not take into consideration the unique characteristics of ideas and expression." 27 This was not merely a formal classification exercise but a substantive judgment that information operates differently from physical products in ways that make strict liability inappropriate.
The Ninth Circuit explicitly declined to create exceptions based on the type of information: "While 'How To' books are a special genre, we decline to attempt to draw a line that puts 'How To Live A Good Life' books beyond the reach of strict liability while leaving 'How To Exercise Properly' books within its reach." 28 The holding was categorical: "Given these considerations, we decline to expand products liability law to embrace the ideas and expression in a book." 29 The court was "[g]uided by the First Amendment and the values embodied therein" in declining "to extend liability under this theory to the ideas and expression contained in a book." 30
While Winter established the general rule against strict liability for information, aeronautical charts carved out what Schultz and Kaner demonstrated was an exceedingly narrow exception.
Abney documented that "[o]ne of the most insidious dangers facing air travellers is an inaccurate aeronautical chart," with "many [aerial crashes] attributed to flight and navigational aids which steered pilots into unmarked obstructions." 31 The two primary suppliers were "the national government and Jeppesen Sanderson, a subsidiary of the Times Mirror Company." 32 Jeppesen's product was distinctive: the company acquired FAA instrument approach procedure data and "portray[ed] the information therein on a graphic approach chart" — this was Jeppesen's "product." 33
Courts held Jeppesen strictly liable on the theory that its charts were products, not mere information — a conclusion driven by the charts' highly specific, operational character. As Kaner documented, "[t]he courts were holding makers of aeronautical charts strictly liable for errors in the charts." 34
The critical insight for defense counsel is that the aeronautical chart exception remained contained. As Kaner's comprehensive review of the case law found, "[t]hrough the 1980s and early 1990s, plaintiff after plaintiff, and article after article argued that aeronautical chart decisions should be extended to other products or that some variation of strict liability should be crafted for situations in which defective information . . ." caused harm — but these arguments consistently failed. 35
Kaner documented that "[i]n most of the cases that I cited . . . plaintiffs lawyers tried — without success — to convince judges to extend strict liability to other information products." 36 His conclusion was definitive: his "reading of the cases, including every published lawsuit that has cited the main aeronautical chart cases and every published lawsuit that has cited Winter, . . ." confirmed Schultz's conclusion that "[a]eronautical charts are the only communication media ever judged by any court [in the United States] to be products" to be products subject to strict liability. 37
Schultz's own exhaustive analysis documented that "numerous courts have considered and rejected publisher liability for defective ideas and information." 38 The wall held: information remained outside the product boundary, with only the most operationally specific, safety-critical navigational charts breaching it.
The question of whether software constitutes a "product" for strict liability purposes remains the most contested boundary in modern products liability law. Lubin framed the challenge directly: the Restatement, "adopted in 1998, . . . admits that as of that time, there were 'no cases on point on their facts.' Instead, the Restatement encouraged judges faced with the question of 'whether to extend strict liability to computer software' to draw their analogies from the 'treatment of software under Article 2 of the Uniform Commercial Code.'" 39
The Drug & Device Law Blog's fifty-state survey confirmed the void: "To date, there have been no reported cases holding a software manufacturer strictly liable for defects in the software." 40 This absence of direct precedent means the law remains fluid, with courts drawing analogies from adjacent doctrines — UCC Article 2, the information products line, and the services/products distinction.
The Reed Smith/Lexology analysis captured the traditional position succinctly: "[f]or decades, product liability law has drawn a bright line between tangible products — think hip implants, pacemakers and insulin pumps — and intangible services like websites and video games." 41 This distinction "historically shielded software developers and digital platform providers from the strict liability claims that have long haunted traditional product manufacturers." 42
The prevailing judicial rationale has been "that software is intangible, often licensed rather than sold, and its defects tend to resemble service failures or contract breaches rather than manufacturing or design flaws." 43 This view, "reinforced by the Restatement (Third) of Torts definition of 'product' as tangible personal property, has led many courts to dismiss product liability claims against software providers." 44
And yet, "[c]ourts have traditionally been reluctant to treat software as a product for purposes of strict liability" — the word "traditionally" carrying an increasingly heavy load. 45
As one Drug & Device commentator put it, "AI doesn't fit neatly into existing legal frameworks. Product liability law, for example, has traditionally dealt with physical products." 46 The shift is particularly visible in the embedded software context: Lubin noted that the "perception is particularly evident when software is embedded in everyday items, from toasters to security cameras, where the software's functionality is integral to the product's operation." 47
The Drug & Device Blog's "Product Line Is Moving" analysis documented that "[r]ecent decisions show that courts are increasingly willing to entertain product liability claims against software-based systems — either in whole or in part — under a few different theories." 48 This represents a genuine doctrinal evolution: "as software and artificial intelligence have become embedded in everything from diagnostic tools to ride-hailing apps, courts are rethinking that approach." 49
Lubin addressed the SaaS dimension directly: "By contrast, SaaS operates on a subscription model, where users pay regular fees to access software hosted in the cloud, with data often managed on remote servers." 50 He declined to distinguish between delivery modes for liability purposes: "I do not distinguish between 'software as a product' (SaaP) and 'software as a service' (SaaS)." 51 His rationale was market-facing: "[r]egardless of these differences, both SaaP and SaaS solutions are regularly referred to as 'products.'" 52
The RAND Corporation's liability analysis highlighted the jurisdictional fragmentation: "How AI-caused harms are treated will differ from jurisdiction to jurisdiction, and such variations could significantly affect the imposition of liability and/or potential size of liability judgments against AI developers." 53 The RAND researchers concluded that "although setting common rules would make it easier for AI developers to manage their risk and make clear to developers and users when they would be held liable, the relative youth of AI technology means that it might not yet be possible to know what these ideal rules should be." 54
Lubin's survey found that "[w]hen this Essay went to press, forty-six of the fifty states (92%) had adopted strict product liability, . . ." but states diverge sharply on whether software falls within their product definitions. 55 The Restatement itself "embraces the majority view of courts that have extended strict product liability to defects associated with electricity once it enters the customer's meter and premises, . . ." suggesting some willingness to extend beyond traditional tangibility — but treating software as analogous to electricity remains contested. 56
The Garcia case represents the most significant judicial engagement with AI product classification to date. The complaint alleged that Megan Garcia's son "became addicted to the [Character.AI] app," resulting in mental health problems, and that the boy fell in love with a chatbot character that imitated Daenerys Targaryen from Game of Thrones. 57 Moments before taking his life, the chatbot allegedly told him to "please come home to me as soon as possible, my love." 58
Garcia alleged "product liability, negligence, intentional infliction of emotional distress (IIED), unjust enrichment, wrongful death, and violations of the Florida Deceptive and Unfair Trade Practices Act." 59 The defendants "initially responded with a motion to dismiss the case on the grounds that the Character.AI chatbot is a service that produces ideas and expressions protected as speech under the First Amendment." 60
Judge Conway rejected this defense, ruling that the Character.AI chatbot is a product rather than a service and declining to dismiss on First Amendment grounds. 61 The court drew on Florida precedent where a trial court had concluded that a ride-hailing "application [wa]s a product under Florida law for purpose of [the plaintiff's] product liability claims" because the claims "ar[ose] from" the app's operational function. 62
The ruling also addressed Google's potential liability as a component supplier, finding that Google Cloud infrastructure access "was necessary to building and maintaining [Character Technologies'] products" and that plaintiff's allegations could "support a plausible inference Google possessed actual knowledge that Character Technologies was distributing a defective product to the public." 63
The AEI's Clay Calvert mounted the most forceful scholarly critique of the Garcia ruling's First Amendment analysis. His starting question was direct: "When artificial intelligence chatbot characters communicate with you through words — when they respond with comments, answers, and questions to your input — are they engaging in 'speech' within the meaning of the First Amendment?" 64
Calvert argued that Judge Conway was "not prepared to hold that Character.AI's output is speech" — a position he found both surprising and unfortunate. 65 He grounded his argument in existing Supreme Court precedent: "Under the US Supreme Court's reasoning in Brown v. Entertainment Merchants Association that gave First Amendment shelter to interactive video games as speech involving a 'virtual world,' chatbot output is speech because it 'communicate[s] ideas' and embraces 'familiar literary devices.'" 66
This argument maps directly onto the Winter framework: if chatbot output communicates ideas and expression — the very categories Winter excluded from products liability — then treating it as a "product" for strict liability purposes could collide with First Amendment protections, just as the Winter court anticipated when it was "[g]uided by the First Amendment and the values embodied therein." 67
An alternative non-product defense frames AI systems as services rather than products. As the Conway order itself noted, defendants argued the chatbot output was a service. 68 Lubin documented that "[j]udicial interpretations have largely upheld this view by treating software as a service rather than a product," noting that "services — from hairstyling through childcare to accounting and consulting — even when provided commercially, are not subject to strict product liability regimes" to strict products liability. 69
The emerging scholarship confirms the significance of this framing. As Browning noted in assessing liability for AI-generated content, "a void in the scholarship exists with respect to the question of who should be liable for AI-generated content," with courts now "confronted with novel legal arguments as they navigate applying traditional legal concepts to disputes involving generative AI." 70
The RAND analysis confirmed the uncertainty: "It is unclear whether courts will define AI as a product, and different products liability tests could lead to very different analyses of whether an AI is truly defective." 71 Plaintiffs "might bring cases alleging that products involving AI are defective; these cases would be decided under specialized products liability law" — but only if the threshold product classification is met. 72
The European Union has taken a fundamentally different approach. As Reed Smith documented, "The EU has adopted Directive 2024/2853 (the 'Product Liability Directive' or 'PLD'), which will take effect on December 9, 2026." 73 This directive "represents a fundamental overhaul of the EU's product liability regime, with far-reaching consequences for technology companies, software developers, and any business placing digital products — including software, AI systems, and digital services — on the European market." 74
Reed Smith identified the most significant change: "the explicit inclusion of software — whether embedded, stand-alone, or delivered as a service — within the definition of a 'product.'" 75 The Directive "clarifies that software is a product for the purposes of no-fault liability, regardless of how it is supplied or accessed (e.g., device storage, cloud, or SaaS models)." 76
This represents a direct legislative rejection of the tangibility premise that has anchored U.S. non-product defenses. Lubin framed the contrast: his essay "contrasts U.S. law with the 2024 European Union Product Liability Directive, which redefines software and artificial intelligence as products, establishing a forward-looking framework that imposes strict liability for software defects." 77
The EU PLD extends beyond classification to create new liability risks. Non-compliance with cybersecurity requirements or failure to provide security updates can constitute a product defect. 78 The PLD and the Cyber Resilience Act "impose ongoing obligations to provide software security updates throughout a product's life cycle, . . ." meaning "[t]he failure to update software or address known vulnerabilities can therefore expose companies to strict liability if a cyberattack exploits an unpatched vulnerability and causes injury or property damage." 79
Procedurally, the PLD "introduces several procedural changes that make it easier for claimants to bring and succeed in product liability claims involving software and cybersecurity, . . ." including "[r]ebuttable presumptions." 80 If a company "cannot demonstrate compliance with these processes, courts may presume defectiveness or causation in favor of the claimant — especially in technically complex cases involving digital products or AI." 81
The EU PLD matters for U.S. defense practice for two reasons. First, it demonstrates that a major jurisdiction has concluded that the tangibility requirement is an anachronism — a legislative judgment that U.S. plaintiffs' counsel will cite as persuasive authority for extending product classification to software. Second, multinational AI developers now face a regulatory environment where the same software is simultaneously a "product" (in the EU) and potentially "not a product" (in the U.S.), creating compliance complexity that defense counsel must navigate.
Drawing from the doctrinal landscape mapped in Sections 2–6, defense counsel confronting product liability claims against AI systems, software, or digital output should construct a layered defense built on five interconnected arguments:
Argument 1: The Tangibility Bar. The Restatement (Third) § 19 defines a product as "tangible personal property." Software, AI output, and digital content are intangible. The Drug & Device Blog's survey confirmed that "many state product liability statutes and other non-restatement-based product liability authority consider 'products' to be limited to tangible things." 82 Defense counsel should establish that the AI system at issue produces intangible output — text, recommendations, images, or interactive responses — not tangible objects.
Argument 2: The Winter Information-Products Defense. For AI systems that generate informational or expressive content, Winter provides direct precedent. The Ninth Circuit held that products liability law "is geared to the tangible world" and declined "to expand products liability law to embrace the ideas and expression in a book." 83 AI-generated text, recommendations, and interactive dialogue are ideas and expression — precisely the categories Winter excluded.
Argument 3: The First Amendment Shield. For AI chatbots and generative text systems, Calvert's Brown v. Entertainment Merchants Association argument provides constitutional grounding. If interactive video games constitute protected speech because they "communicate ideas" and use "familiar literary devices, including characters and dialogue," then chatbot output — which communicates through words, narrative, and dialogue — warrants equivalent protection. 84 The Winter court was explicitly "guided by the First Amendment" in declining to extend strict liability to informational content. 85
Argument 4: The Services Classification. Where software operates through subscription, cloud delivery, or interactive user engagement (SaaS, API access), defense counsel should argue it is a service, not a product. Judicial interpretations "have largely upheld this view by treating software as a service rather than a product." 86 The licensing-not-selling distinction remains relevant: software "is intangible, often licensed rather than sold, and its defects tend to resemble service failures or contract breaches." 87
Argument 5: Jurisdictional Specificity. Given the "widespread doctrinal inconsistencies within the common law" documented by Lubin's fifty-state survey 88, defense counsel must ground their arguments in the specific jurisdiction's treatment of the product definition. The absence of any reported case "holding a software manufacturer strictly liable for defects in the software" 89 means this remains an area where persuasive authority, policy arguments, and analogical reasoning carry exceptional weight.
Defense counsel should be prepared for three principal counterattacks:
The Garcia Precedent. Plaintiffs will cite Conway's ruling that a chatbot is a product. Defense counsel should emphasize that Garcia is a single district court ruling on a motion to dismiss — not a merits determination — and that it applied Florida law in a factual context involving a minor's suicide, creating enormous sympathetic pressure.
The EU PLD as Persuasive Authority. Plaintiffs will argue that the EU's explicit inclusion of software as a product reflects the "modern" view. Defense counsel should respond that the EU PLD is a legislative choice reflecting a different legal tradition — European strict liability has always been broader — and that it confirms the current state of U.S. law requires a similar legislative act to change the product boundary.
The Embedded Software Analogy. Plaintiffs will argue that software embedded in physical products (medical devices, autonomous vehicles) is already treated as part of the product, and that standalone software should be treated equivalently. Defense counsel should distinguish embedded software — which is integral to a tangible product's function — from standalone software, AI output, and cloud-delivered services, which lack any tangible product nexus.
The most powerful strategic lesson from the aeronautical chart cases is what Kaner and Schultz demonstrated: even when courts carved out a narrow exception for highly specific, safety-critical navigational products, that exception did not spread. For over three decades, "plaintiff after plaintiff, and article after article argued that aeronautical chart decisions should be extended to other products" — and failed. 90 Aeronautical charts remain "the only communication media ever judged by any court [in the United States] to be products" to be products subject to strict liability. 91
Defense counsel should invoke this containment principle: even if Garcia or future rulings classify specific AI applications as products, the history of the aeronautical chart exception demonstrates that courts have consistently resisted expanding narrow product classifications into broad categorical holdings. The non-product defense, properly articulated, remains the strongest gateway defense available to AI developers and software publishers facing strict liability claims.
This module synthesizes evidence from the following 41 warehouse sources: