The primary obstacle to the widespread adoption of agentic commerce is not a lack of technical capability, but a fundamental misalignment of economic risk. Merchants are currently being asked to bear the full financial liability for errors made by autonomous agents — software they do not control, cannot audit in real-time, and for which no clear dispute resolution framework exists. When an agent misinterprets a prompt or selects the wrong SKU, the merchant is the party left holding the bill.
The 93% Disconnect
This liability mismatch is creating a significant drag on market participation. According to PYMNTS Intelligence data from September 2026, 93% of merchants believe that AI or agent providers should bear the financial loss when an agent makes an incorrect purchase. Faced with this exposure, merchants are acting with predictable caution: only 28% of those surveyed are willing to offer their full product range to AI agents under current terms. Instead of embracing the technology, most retailers are effectively self-insuring by limiting their inventory exposure to mitigate the risk of unpredictable software behavior.
The Chargeback Vacuum
The current payment infrastructure is ill-equipped to handle these disputes. The existing system was designed to adjudicate disagreements between humans and merchants, not to parse the intent of a generative model. There is no clean reason code to distinguish an “agent error” or misunderstanding from a legitimate, cardholder-authorized purchase. Because the platforms enabling these agents are not currently liable under any network rule, the merchant absorbs the cost of the dispute by default. This creates a structural vacuum where the party best positioned to prevent the error — the AI provider — has no financial incentive to do so.
The Regulatory Void
The legal landscape offers no refuge for merchants. The Electronic Fund Transfer Act and Regulation E were written for a world of human-initiated transactions, and as of late 2026, there is no federal framework that specifically addresses agent-delegated purchases. It remains legally unresolved whether granting an agent access to payment credentials constitutes valid consumer authorization under existing law. With the CFPB and the Federal Reserve remaining silent and the current administration signaling a preference for innovation over immediate regulation, the industry is left to navigate a landscape without a clear rulebook.
The Patchwork of Solutions
Card networks have begun to acknowledge the problem, but their responses remain fragmented and incomplete:
- American Express: The network is currently the only one attempting to address this directly with its Agent Purchase Protection, announced in April 2026. However, this protection is conditional, requiring registered agents and authenticated purchase intent, with the necessary technical specifications for “Cart Context” still under development.
- Visa and Mastercard: Their respective TAP and Agent Pay frameworks provide the signaling necessary for a merchant to identify that an agent is initiating a transaction. Crucially, these protocols do not re-allocate liability for non-fraudulent errors.
- EMVCo: The draft agentic payments framework published in September 2026 leaves the core question of liability allocation for non-fraud disputes unresolved.
The Binding Constraint
Technical barriers are falling. With the October 2026 launch of Constructor’s Stripe-powered agentic checkout, retailers such as Sephora, The Very Group, Monica Vinader, White Stuff, and Furniture Village now have the technical capability to enable agentic purchasing. The “welcome” posture from major retailers is becoming actionable.
However, the liability question remains the binding constraint. While identity verification and transaction scoring — such as the Mastercard probability score announced September 30 — are essential for preventing fraud, they do not solve the economic risk of a misunderstanding between an agent and a merchant. As Federal Reserve Governor Christopher Waller noted in his September 2026 speech at Sibos, trust is a market-structure problem. Until there is a standardized way to allocate the cost of agent error, the growth of agentic commerce will remain constrained not by what the technology can do, but by what merchants are willing to risk.