AI Product Liability

AI litigation is increasingly centered on a familiar area: product liability. Early cases are examining whether AI applications used by consumers are classified as products or services, and whether the alleged harms stem from design flaws, insufficient warnings, or predictable misuse. This change is supported by lawmakers, including the EU’s updated directive on liability for defective products (PLD), as well as a growing number of US state laws. All of these developments suggest that product liability will play a key role in the upcoming wave of AI-related legal cases.

It's interesting to see how many of the initial headlines about AI disputes focused on issues like consumer protection, privacy, defamation, and intellectual property. However, product liability takes a different approach—checking mass-produced technologies for defects, warnings, and what could have been foreseen. This liability can involve everyone involved in bringing the product to market. As AI becomes more integrated into our daily lives, people are increasingly likely to view AI-powered experiences as products that can be held responsible for.

A common challenge in these disputes is how courts should view generative AI outputs. Defendants suggest that chatbot responses are a form of expressive content, thereby framing the case as one of speech liability rather than product design. On the other hand, plaintiffs focus on the system's architecture, such as guardrails, defaults, escalation pathways, and marketing. This perspective tends to make the case resemble a product defect issue more closely.

In many cases, people tend to see the AI system as the main experience users interact with. This viewpoint helps prevent disagreements over whether a particular output constitutes protected speech. Instead, it encourages looking at the system’s design choices as potential issues. It also makes it possible to hold various parties involved in putting the system into use accountable, including those behind branded apps and other contributing actors.

The product-liability doctrine applies to technologies that are widely distributed and provide consistent experiences for users. As courts consider whether certain AI applications should be viewed as products or services, plaintiffs are increasingly leaning on familiar product liability claims. These include design defects—such as guardrails, interaction design, and safety features—failure to warn of limitations or potential misuse, and negligence, which involves reasonable testing and ongoing monitoring.

Another common theme involves supply-chain liability, with early rulings indicating that plaintiffs are open to testing theories that go beyond the model developer. They consider the company that brands and uses the system, as well as upstream providers who may have helped or significantly contributed to the final product’s integration. Legislation like California’s AB 316 highlights a growing trend toward making causation disputes more fact-specific, emphasizing the importance of detailed, case-by-case analysis.

Policy developments are adopting the terminology of the products doctrine. This matters because these frameworks influence strategies and can lend persuasive authority to arguments about defect, foreseeability, and standard of care, even when claims are grounded in common-law tort.

Several recent developments signal a noticeable shift in how AI systems are viewed and regulated. The EU’s Product Liability Directive (PLD) treats software, including AI systems, as products and applies strict liability principles across the entire distribution chain. This includes parties that significantly modify AI systems. Member states are expected to adopt the directive by December 2026.

In the United States, a Senate proposal echoes a similar emphasis on product liability for certain AI systems, though it isn’t yet finalized. At the state level, laws such as California’s AB 316 and SB 243 are often invoked by plaintiffs to argue about foreseeability and to define what safety features are reasonable in specific situations.

The EU framework’s influence extends beyond Europe. Its concepts are likely to be referenced in US lawsuits and expert reports, especially when companies market a single AI-enabled product across multiple jurisdictions. Additionally, detailed state laws can serve as important signals in shaping tort standards. cases.

What This Means

Several themes are shaping the next wave of developments. Courts will continue to explore the distinction between product and service, which can affect how strict liability theories are applied and how warnings and designs are assessed. We're seeing more focus on AI architecture in legal pleadings, rather than just on individual outputs. At the same time, liability theories are evolving along the AI supply chain, as plaintiffs look into possibilities involving different actors both upstream and downstream. Lastly, regulation is moving toward a shared liability approach: documents like the PLD and specific state laws are likely to be cited in complaints and expert reports to discuss defect and foreseeability. Meanwhile, testing artifacts, monitoring signals, and change histories remain crucial during discovery, helping shape causation narratives and settlement discussions.

Two essential disciplines are consistently important in product cases: delineating the product and documenting the design narrative.

• Charting the deployed system helps prevent ambiguity about the product's state at any given time, especially when behaviors fluctuate with updates.

• Simultaneously, maintaining up-to-date documentation of testing, risk assessments, and safety trade-offs frequently forms the basis of defect and foreseeability assertions; it serves as the record that permits a defendant to explain not only what was constructed but also why the design decisions were deemed reasonable at the relevant time.