.png)
Nobody Owns the Trust Layer: The Accountability Gap Slowing Agentic Payments
Meta title: Agentic Payments: Who Owns the Trust Layer?
Meta description: Agentic payments are advancing, but accountability is still fragmented. See what merchants need to define around mandates, budgets, evidence and reconciliation.
Agentic payments are moving from demonstrations towards real transactions, but the commercial trust layer remains unclaimed. The problem is not simply whether an AI agent can pay. It is whether every party can show what the agent was authorised to do, identify who was responsible for its actions, and resolve the transaction when something goes wrong.
This week’s signals capture the tension. Stripe president John Collison said agentic commerce “started going vertical” in September, but shared no figures. A separate report put Stripe Link agent transactions up 38 times and payment volume up 20 times in September, again without absolute totals. At the same time, banks are asking how an agent’s authority can be bounded and attributed, while regulators are emphasising oversight and risk controls.
For merchants, the implication is practical: trust is not a feature a provider can simply switch on. It is a contract and operating model, backed by clear mandates, spending limits, evidence and reconciliation.
Growth claims are not the same as accountable commerce
On the 12 October Cheeky Pint podcast, Collison said Stripe’s agentic commerce activity had “started going vertical” in September. The comment points to momentum, but without transaction counts, payment values or a comparison period, merchants cannot assess the size or quality of that growth.
Reporting by Forkast notes that Stripe’s September 2025 Instant Checkout launch with OpenAI was retired by March 2026, after near-zero sales from about a dozen Shopify merchants. That is a reminder that a live integration or a promising trend line does not, on its own, prove repeatable demand.
Meanwhile, The Decoder’s account of Link data reports 38-fold transaction growth and 20-fold payment volume growth in September, but says the underlying totals were not disclosed. Growth rates are useful signals. Without a disclosed base, they do not tell an operator how much commerce is taking place or how much operational exposure it creates.
The trust question is just as important as the volume question. Forkast cites Product.ai data that only 14% of consumers trust AI to execute purchases without manual verification. It also cites September 2026 PYMNTS Intelligence research finding that 93% of merchants believe AI or agent providers should bear the financial loss when an agent makes an incorrect purchase. Those figures point to a mismatch: consumers want control, and merchants want someone else to take responsibility when the agent’s decision causes a loss.
The missing layer is proof of authority
An approved payment credential does not necessarily prove that a particular purchase was authorised. A card or token can be valid even if the agent selected the wrong item, exceeded a budget, misunderstood a constraint or acted on instructions that were no longer current.
That distinction is becoming central to AI payments. In PYMNTS’ reporting on FIS, FIS executive Troy Mann argues that intent must be expressed and captured. For merchants and their payment partners, this means being able to connect a transaction to the instruction, limits and approvals that were in force when the agent acted.
Banks are approaching the same issue through bounded autonomy and accountability. An e27 report on a UOB report describes the industry’s shift from “know your customer” towards “know your agent”. Identifying the customer is still essential, but operators also need to know which agent acted, who operates it and what authority it was given.
The Financial Stability Board’s 15 October final report sets out 12 sound practices for responsible AI adoption in financial institutions. One point is particularly relevant to commerce: “AI agents do not reason like humans and may interpret pre-defined goals in unexpected ways”. Human oversight and controls should therefore reflect the potential impact of a decision, not just the apparent confidence of the system.
Payment tools are advancing faster than shared responsibility
The payments industry is building ways for agents to transact. Mastercard’s Agent Pay for Machines entered early access on 20 October, with a demonstration involving nine payments to two merchants within one budget. The approach uses signed payment vouchers when a service requests payment. Mastercard also announced a Stablecoin Security risk-assessment service this week.
These developments show that the technical building blocks are progressing. They do not, by themselves, establish a shared process for deciding who bears a loss when an agent acts outside its mandate or the transaction record is incomplete.
This is why protocols and payment tools can be identity-first and money-later: participants need confidence in the authority and evidence attached to the transaction before they can agree on how it should be funded or disputed. Merchants should track these changes, but avoid treating integration availability as a substitute for clear commercial terms.
Quantum Payments has also examined how payment control and agent activity affect merchants. For related context, read our analysis of why banks have principles while merchants face disputes, how agent payments are being measured, and the movement of AI agents through the money layer.
An operator checklist for accountable agentic payments
Before enabling agent-led purchasing, merchants and payments operators should make accountability tangible in their systems and agreements.
Frequently asked questions
What are agentic payments?
Agentic payments occur when an AI agent initiates or completes a payment on behalf of a person or business, within whatever authority and controls have been provided.
Why is accountability still unclear?
The transaction may involve several parties: the customer, agent provider, merchant, payment provider and financial institution. Identity, instructions, transaction evidence and responsibility can be held across different systems and agreements.
What should merchants do first?
Start by documenting the authority granted to agents and the limits on that authority. Then confirm that payment records, contracts and reconciliation processes can show what happened and who is responsible for responding.
Sources