The Argument
When an agent makes a purchase that the human did not intend, who is liable? When an agent negotiates terms that disadvantage its principal, who bears the loss? When a merchant delivers a product that meets the agent's specifications but not the human's expectations, whose fault is it? These are not edge cases. They are the structural liability questions that will define the legal architecture of agentic commerce. Current consumer protection law was built for a world where the customer and the decision-maker are the same person. In agentic commerce, they are not. The essay argues that the liability gap - the space between existing legal frameworks and the reality of agent-mediated transactions - is the most urgent governance challenge in the agentic economy.
The Evidence
The essay maps five liability domains. Dispute resolution architecture must handle a new category of disputes: the agent acted within its technical scope but outside its principal's intent. Traditional chargeback and refund mechanisms assume the customer made the decision. When the agent made the decision, the dispute resolution process must determine whether the agent was properly delegated, whether the delegation was current, and whether the agent's interpretation of its mandate was reasonable.
Accountability surfaces must be designed for multi-party transactions where the human principal, the agent platform, the agent itself, and the merchant all bear partial responsibility. Consumer protection in agent-mediated transactions faces a fundamental question: does the Consumer Rights Act apply when the "consumer" is a machine? The essay examines how existing regulations - including the UK Consumer Duty, the EU AI Act, and the US FTC guidelines - apply (or fail to apply) to agent-mediated commerce. Insurance and risk transfer models must evolve to cover agent-specific risks: delegation drift, authority expiry, and the gap between agent capability and agent authorisation. The essay introduces the concept of liability architecture - the designed distribution of accountability across all participants in an agent-mediated transaction.
The Implication
Liability architecture is not a legal afterthought. It is a design requirement. Every agent-mediated transaction creates a liability distribution - whether that distribution is designed intentionally or emerges by default. The organisations that design their liability architecture proactively - defining clear accountability surfaces, building machine-readable dispute resolution processes, and establishing verifiable delegation records - will operate with legal clarity. Those that allow liability to emerge by default will face regulatory exposure, consumer harm, and commercial disputes that their existing legal frameworks cannot resolve. The liability gap is closing. The question is whether it closes through design or through litigation.