Should AI Companies Be Liable for Their Models' Harms?

“Product liability law already applies to defective software in medical devices and autonomous vehicles. There is no principled reason that an AI chatbot providing suicide instructions to a depressed teenager should enjoy immunity that a medical device manufacturer does not. The harm is real, the causal chain is traceable, and liability law exists precisely to internalize these costs.”

“The EU's experience shows that liability-driven compliance is achievable. Companies operating in Europe have built audit trails, redress mechanisms, and human oversight systems that their US equivalents have not — demonstrating that legal pressure produces safety investment rather than stifling it.”

“The documented harms of AI are real but not unprecedented. Cars kill 40,000 Americans per year; social media has documented mental health effects; pharmaceuticals cause adverse events. The existing tort system handles these categories. AI-specific liability reform requires showing why existing law fails here, not just that harms exist.”

“Strict liability for AI outputs is economically equivalent to strict liability for the printed word: publishers could be sued for what their authors write. The liability framework must distinguish between negligent safety practices and the inherent unpredictability of stochastic systems. Conflating the two is not law; it is policy by litigation.”