No Rule Yet Verifies the Human Behind an AI Agent's Payment, Legal Experts Say

The Liability Gap Behind Machine Payments

No rule yet tells a merchant how to verify the human behind an AI agent's payment, and that verification gap is the core of the current identity debate. AI agents already pay for data, computing power and web searches on their own, while the law still assumes a person approves every transfer. The specialists at the center of the debate point to blockchain as the layer that could close the hole.

The market itself is still small. A research report tracking agentic payments recorded 6.4 million x402 payments between July 23 and August 26, 2026, most of them tiny sums for data feeds and API calls. x402 is a standard that lets software buy paid web resources without human sign-off, the kind of routine machine-to-machine commerce often grouped under PayFi. The dollar volumes are modest, but the pattern matters: machines initiating payments at scale is no longer theoretical.

Banks are not waiting for clarity. JPMorgan now blocks outside AI agents from customer accounts, and CEO Jamie Dimon has pointed publicly to the risk of agents going rogue. Without a verification standard, an outside agent looks to a bank like an uncontrolled party with account access.

US statutes offer merchants little guidance. The Electronic Fund Transfer Act, written in 1978 for a world in which a human authorized each transfer, still governs unauthorized payments. The GENIUS Act sets rules for stablecoin issuers but says nothing about the agents that spend those coins, a corner of the PayFi market no draft reaches.

The combined effect is an accountability vacuum: nobody can say for certain who pays when an agent books the wrong flight. That is a question of identity as much as liability, since liability law needs a person to hold to account and an unverified agent offers none.

Zero-Knowledge Proofs and Europe's Disclosure Rule

Vyara Savova, a senior policy expert at the European Crypto Initiative, expects most people to hand routine payment tasks to agents. In her view, a single chain can bundle what those agents lack: the payment rails, an identity layer of the kind blockchain domains already provide for naming, and a trust layer. “So then we have the payment rails, the identification rails, basically the trust layer, all in one, which would to a very large extent solve this,” she said.

Lawyer Anja Blaj Zajc rejects the idea that users must choose between privacy and accountability. A zero-knowledge proof confirms a fact without revealing the data behind it, and in her reading it removes that choice. “You would have the ZK proof that proves that there’s indeed a human behind that agent, that there’s going to be somebody responsible for that agent,” Blaj Zajc said.

The point carries legal weight. A ZK credential could tie an agent to a responsible owner without exposing that owner to every merchant it pays. Liability law needs a person to hold to account; an anonymous wallet offers none.

Lord Chris Holmes, a Conservative member of the UK House of Lords who sponsored a private member’s bill on AI regulation, puts disclosure first. “It always needs to be clear AI was used, how it was used, for what purpose, and that consent and trust and transparency were in that mix,” Holmes said.

Europe already enforces part of that. Since August 2, 2026, Article 50 of the EU AI Act has obliged chatbot providers to tell users they are dealing with an AI system, and national authorities can enforce the rule. The AI Omnibus pushed high-risk deadlines to 2027 and 2028 but left this duty standing. The Act covers disclosure, not payments.

The Question Merchants Still Cannot Answer

Our reading of the two strands is that they converge on the same missing primitive: a verifiable link between an agent and an accountable human. The most load-bearing primary document in this debate, the text of the EU AI Act, states a disclosure duty for chatbot providers and nothing on payment verification, and no US statute in the record answers the identity question either. Among the proposals on record, the ZK credential is the only one that answers privacy and liability at once, which is why the legal specialists in the debate back it. Until a rule tells a merchant how to verify the human behind a paying agent, banks can be expected to keep their doors shut; that rule is the single piece that would close the gap.