AI agents are on track to handle $8 billion in commerce by 2026, a projection that underscores the rapid adoption of autonomous payment systems. But as these digital workers take over transactions, regulators and businesses are grappling with a critical flaw: no one is clearly responsible when something goes wrong.
The Rise of Agentic Payments
Agentic payments refer to transactions initiated and executed by AI software without direct human intervention. Think of a virtual assistant that books a flight, pays for a subscription, or negotiates a price—all on its own. The $8 billion forecast highlights how fast this is becoming normal. Companies are deploying agents to cut costs and speed up processes, from supply-chain settlements to customer refunds.
But the convenience comes with a catch. When an agent pays the wrong vendor, overcharges a client, or falls for a scam, traditional payment rails offer little recourse. The transaction happened, and the money moved. Unwinding it is messy, and assigning blame is even messier.
Who's Liable When an Agent Goes Wrong?
The accountability gap is the core of the risk. If a human employee makes a payment error, the company is liable. If a software agent does the same, the law is often silent. Is the developer at fault? The company that deployed the agent? The platform it ran on? Regulators haven't staked out a clear position, and firms are left guessing.
That uncertainty matters. A faulty agent could trigger a cascade of unauthorized charges, and the victims—whether consumers or other businesses—may have no clear path to a refund. Fraudsters are already eyeing agentic systems, knowing that the lack of accountability makes it harder to trace and recover stolen funds.
What Regulators and Firms Face
For regulators, the challenge is keeping up. Payment rules were built for humans pushing buttons, not AI acting autonomously. Questions about licensing, supervision, and consumer protection are all up in the air. Firms, meanwhile, have to decide how much risk they're willing to absorb. Some are building in guardrails—like spending limits and human-in-the-loop approvals—but those add friction and defeat part of the agent's purpose.
The stakes are high. $8 billion in commerce is still small compared to overall payment volumes, but the trajectory is steep. If accountability isn't sorted soon, a major incident could set back adoption for years.
Who pays when an AI agent messes up? That's the question regulators and companies still can't answer.
