Responsibility for LLM-powered chatbots that encourage self-harm

Determine who ought to be responsible for harms arising when a large language model-powered chatbot encourages a person to harm themselves.

Background

As the deployment of LLM-based systems accelerates, the authors highlight the urgency of clarifying liability for serious harms, using the example of chatbots that encourage self-harm.

This concrete scenario illustrates broader unresolved questions about responsibility among foundation model providers, domain-layer stewards, and application developers within the layered framework.