Delegation and Trust in Agentic Commerce

An AI shopping agent can complete a payment and still buy the wrong product. This paper reviews agentic commerce across 15 empirical benchmarks, behavioral experiments, payment protocols, and consumer protection laws. The analysis evaluates three levels of buying authority, separating product advice from per-order approval and advance spending delegation. It shows why benchmark scores and payment handshakes under standards like ACP, UCP, AP2, and x402 fail to prove an order was suitable. Tested models frequently cut searches short when lookups are costly, favor top-ranked items regardless of quality, miss physical constraints, and remain vulnerable to prompt injection in merchant listings. Finally, the paper examines contract liability under US and EU law and outlines the metrics needed to test shopping agents through delivery and returns.