Source: CEPA

A fictional user named “Alex” is portrayed as suffering worsening mental health after prolonged interaction with a generative AI chatbot, prompting the question whether the chatbot’s developer should pay compensation. Real-world reports cited in the piece allege chatbots have reinforced delusional beliefs or provided unreliable financial and health advice, but responsibility is unclear across the US, Europe, and the UK. All three jurisdictions recognize negligence principles, yet plaintiffs face major evidentiary obstacles: obtaining internal technical information from AI companies, proving causation between model outputs and harms, and defining what constitutes a “safe” or “reasonable” chatbot behavior given AI’s novelty.

The European Union’s 2024 update to the Product Liability Directive to include software and AI could allow plaintiffs to seek redress through member-state rules or the bloc’s regime, with courts potentially ordering disclosure of technical material or shifting the burden of proof to defendants in complex cases. EU product liability covers limited harm types and may require showing a medically recognized psychological illness, while the bloc also offers a class action mechanism to share costs. In the US, no federal product liability law exists and state rules vary; recent litigation such as Megan Garcia v Character A.I. has seen an interim order treating a chatbot as a “product,” and the US’s strong class action system makes collective suits feasible despite similar causation and standard-of-care challenges. The UK’s Consumer Protection Act 1987 still targets tangible objects and has not been updated to cover software or AI, leaving plaintiffs there with comparable hurdles unless common-law judges take novel approaches in future cases.