AI agents are starting to buy things, but who handles the dispute when they get it wrong?
Chargebacks911, a global leader in dispute resolution and chargeback prevention, is warning that the payments industry is building agentic commerce from the wrong end. As Visa, Mastercard, and American Express announce frameworks, programs and live pilots for AI-initiated transactions, Chargebacks911 says the post-transaction infrastructure needed to manage disputes, assign liability, and resolve contested charges in a world without a human buyer remains almost entirely unaddressed.
The warning follows a series of major industry moves in recent months. Visa expanded its Agentic Ready program globally in late April 2026. Mastercard and Santander completed Europe’s first live end-to-end AI agent payment within a regulated banking framework in March. American Express introduced an agentic commerce developer kit alongside a commitment to cover erroneous purchases made by registered agents on its network. The infrastructure for agentic transactions is being built at speed but the infrastructure for agentic disputes is not.
The stakes are significant. According to Mastercard’s 2025 State of Chargebacks report, based on research from Datos Insights, global chargeback volume is already forecast to grow 24% from 2025 to 2028, reaching 324 million transactions annually – before the full impact of agentic commerce is factored in. The Consumer Bankers Association has warned separately that if AI agents make mistakes that generate widespread disputes, ordering wrong products, duplicating purchases, or acting on ambiguous instructions, that volume could overwhelm existing dispute resolution infrastructure entirely.
The gap matters because disputes in agentic commerce are structurally different from anything existing chargeback frameworks were designed to handle. Traditional dispute resolution rests on a single foundational assumption: a human being made a decision. Intent, authorisation, and liability are all determined by reference to what a cardholder chose to do. When an AI agent makes the purchase autonomously, acting on delegated authority, within parameters set at a prior point in time, without explicit confirmation at the moment of transaction, that assumption collapses.
“The card networks have built dispute frameworks over decades around one idea: the cardholder either did or did not authorize this transaction. In an agentic world, that question doesn’t have a clean answer,” said Monica Eaton, founder and CEO of Chargebacks911. “Who authorised it? The consumer gave the agent permission to act. Did that permission cover this specific transaction? That depends on what the agent was told, what it inferred, and whether the merchant can prove any of it. But most merchants cannot.”
Chargebacks911 says the problem runs deeper than reason codes. In a human-driven transaction, a dispute can be tested against behavioural signals: did the cardholder’s purchase pattern suggest intent? Does the device and session history match? Is there a history of similar transactions? These signals inform both fraud detection and dispute outcomes. An AI agent disrupts every one of these markers. Its behaviour is consistent, methodical, and often indistinguishable from automated activity, making it difficult for merchants to defend legitimate transactions and equally difficult to identify genuinely disputed ones.
“The industry spent years training fraud and dispute systems to read human behaviour. Agentic commerce doesn’t produce human behaviours but it produces something more consistent, more data-rich, and more auditable, but only if merchants have built the right infrastructure to capture it,” said Donald Kossmann, chief technology officer at Chargebacks911. “The merchants who understand this early will have a structural advantage. Their dispute rates will fall, their recovery rates will rise and their evidence quality will be higher than anything legacy systems can produce.”
Chargebacks911 addresses this through its Unified Dispute Management System, UDMS, which uses AI and machine learning to build and interrogate the evidence architecture that agentic transactions require. Rather than relying on point-of-sale signals or human-centered behavioural data, UDMS captures the full consent and permission trail – what the agent was authorized to do, the scope and limits of that authorization and a timestamped record of each action taken. Mapped across the transaction lifecycle, this gives merchants and financial institutions the visibility needed to accurately classify disputes, defend representments, and identify where agent behaviours fell outside the parameters the consumer originally set.
Chargebacks911 recommends that merchants, acquirers, and financial institutions take three steps ahead of the next wave of agentic commerce adoption. First, establish granular permission and scope frameworks for any AI agent transacting on their platforms or on behalf of their customers, ensuring that authorization is documented at the point of delegation rather than reconstructed after a dispute. Second, build evidence capture infrastructure that logs agent behaviour continuously across the transaction journey, not only at checkout. Third, review existing fraud detection rules and dispute thresholds to account for the behavioural differences between human and agent-initiated transactions, systems calibrated for human behaviour will generate increasing numbers of false positives as agent-led commerce scales.
“Visa, Mastercard and Amex are doing exactly what they should be doing, enabling the front end of agentic commerce to function,” said Eaton. “The question nobody is asking loudly enough is what happens at the back end, when a consumer looks at their statement and doesn’t recognize a charge that their agent authorized three weeks ago. That dispute is coming and the merchants who have built the evidence trail will resolve it in minutes. The ones who haven’t will lose the revenue, pay the fee, and have no way to fight it.”
Chargebacks911 supports clients in 87 countries and safeguards more than 2.4 billion transactions per year through its UDMS and ResolveLab platforms. The company continues to invest in machine learning, automation, and network connectivity to help payments teams manage evolving dispute and fraud risk in an increasingly agentic commerce environment.