USA Trends and Developments Contributed by: Gregory Ulmer, James Phillips and Ryan Walton, BakerHostetler
• The company knew, or should have known, users would rely on outputs. • The system could have been designed to slow, escalate or qualify advice. • The harm flowed not from misuse but from design choice. By summary judgment, the debate is no longer phil - osophical. Discovery has surfaced internal design deliberations, product testing data, and early drafts of safety language. At trial, plaintiffs strip away tech - nical complexity and offer a simple narrative: “They knew people would trust it – and they chose speed to market over safety.” Disclaimers and licensing terms fade into the back - ground. Jurors hear about judgment, priorities and preventable risk. Once reliance is established, dam - ages analysis expands rapidly, moving beyond eco - nomic loss into emotional and moral harm. The AI sys - tem ceases to appear novel; it becomes a tool that should have been safer. Vignette Two : “ This didn ’ t have to be designed this way .” In a second scenario, a consumer-facing AI applica - tion optimises engagement using adaptive algorithms. Safety features – filters, rate limits, escalation prompts – exist, but are disabled by default to reduce fric - tion during onboarding. Internal metrics confirm that engagement drops when guardrails are visible. After a serious harm involving a vulnerable user, plain - tiffs frame the defect not around content but archi - tecture. • Safeguards existed but were buried. • Risk signals were detected but not acted upon. • Engagement metrics were prioritised over interven - tion. During discovery, engineers and product managers testify that increasing friction would have lowered engagement and adoption. From a technology stand - point, these tradeoffs were rational. From a product liability lens, they appear ominous.
At trial, plaintiffs rely on familiar product-case image - ry: internal emails, A/B test results, and side-by-side comparisons of what was shipped versus what was feasible. The defence emphasises unpredictability and user choice. Plaintiffs respond with the most danger - ous sentence in the nuclear verdict lexicon: “They didn’t have to design it this way.” Once the case is framed around preventability, jurors substitute hindsight for engineering judgement. AI becomes legible not as cutting-edge technology but as a familiar product allegedly missing adequate safe - guards. Why these vignettes matter These scenarios illustrate how AI litigation increas - ingly inherits the most volatile characteristics of prod - uct liability cases. Design-defect narratives translate easily. Foreseeable reliance magnifies non-economic damages. Formal defences – service labels, licensing terms, disclaimers – lose practical force long before verdict. Critically, exposure often crystallises early. Motions to dismiss may fail not because courts have embraced a radical redefinition of products, but because plaintiffs have successfully framed the case in terms that juries already understand. Once that framing survives, litiga - tion economics shift dramatically. Strategic implications The lesson is not that courts have definitively clas - sified AI as a product – it is that, in a nuclear-verdict environment, the operative question is whether the case feels like one. When jurors are asked whether harm was foreseeable and design choices reasonable, the gravitational pull of product liability reasoning takes over. At that point, contractual disclaimers and service labels offer limited protection against moral blame. Practical Risk Management and Litigation Strategy Considerations In this environment, effective risk management requires anticipating how design decisions will appear years later under hindsight scrutiny.
331 CHAMBERS.COM
Powered by FlippingBook