Product Liability Foundations
Who answers when AI causes harm? Two regimes: fault liability (prove an action/inaction caused harm) and strict liability (no-fault - prove only defect plus causation). AI breaks these frameworks because attribution is hard and models are opaque; the open question is whether AI will be classified as a "product" at all.
Who answers when AI causes harm? Product liability law holds economic actors → retailers, distributors, manufacturers → responsible for product harm. Two regimes, two AI-shaped problems.
- Fault liability → victims must prove an action or inaction by the product maker caused the harm (e.g. noncompliance with a product safety law, or negligence - failing to exercise due care). The maker must be at fault.
- Strict liability → the no-fault regime; no need to prove wrongdoing, only that the product was defective and the defect caused the harm
Why AI breaks the old frameworks:
- Attribution is hard → AI is autonomous and constantly evolving; ML systems learn patterns independently and generate outputs (like an autonomous vehicle's micro-decisions) without human direction, so harm may stem from behaviour that was virtually impossible to foresee when the system was built.
- Complexity & opacity → deep learning models can be black boxes even to the engineers who built them, let alone courts trying to trace what caused the harm.
Product liability is state-level law, shaped by court decisions, with little AI case law → ambiguity. Three claim types apply broadly → strict liability (prove harm from a defective product) · negligence (failure to exercise necessary care) · breach of warranty (failure to fulfil promises made about a product). The FTC has warned about unsubstantiated accuracy claims for biometric tools like facial recognition. The critical open question → will AI systems and services be classified as "products" at all?
Prepare accordingly → liability reaches third-party vendors plus organisations developing, using and deploying AI, including distributors and importers; litigation may force compensation and disclosure of sensitive AI information, so R&D teams must know the exposure and senior leadership must prioritise safe AI product development.