Agentic commerce chargeback liability: who pays when an AI agent buys wrong
Agentic commerce got its own standards effort for one reason: nobody can prove what an AI agent did. On July 13, 2026, A-Comm Technologies — founded by former Visa payments leaders — opened public comment through August 14 on the A-Comm Evidence Protocol, an open-source standard for tamper-evident records of agent transactions, built around observability, traceability and auditability. It exists because card dispute rules assume a human buyer, and agent-initiated purchases break that assumption: the usual chargeback evidence (IP, device fingerprint, navigation path, time on site) is now generated by software. Regulators are not waiting — the UK's CMA told businesses in March 2026 that if an AI agent you use does something illegal, you are responsible. This is an analysis of where liability actually lands today, why bolt-on evidence layers are needed on reversible rails, and why an agent paying USDC on Base over x402 settles the question differently: the payment is final, non-custodial, and already leaves a receipt anyone can verify.