The question is older than software
When the Bardi family of Florence extended credit to wool buyers in England in the 1300s, they faced the same problem modern payment systems still carry. How do you authorize money to move when the authorizing party is not present?
Their answer was a letter of credit bearing the personal seal of the issuing merchant, sent ahead to a correspondent banker in Venice or Bruges or Constantinople. The seal was not decorative. It was the authorization. It answered the only question the receiving banker needed answered. Who told you to do this?
That question has anchored every payment authorization framework built since. It predates electronic payments by six centuries. It is also the question agentic commerce cannot answer cleanly with the infrastructure we have today.
The law still assumes a person
Follow the lineage. The Electronic Fund Transfer Act of 1978 defined an authorized electronic fund transfer as one initiated by a person with the consumer's consent. In 1980, NACHA built the ACH framework around the same anchor. UCC Article 4A tied wire transfer liability to commercially reasonable security procedures for verifying sender identity. The E-SIGN Act extended that logic into electronic signatures, but the signer was still human.
Every one of these frameworks asks the same question first. Who authorized this? Every one assumes the answer ends with a person whose identity the system can verify and whose liability a court can assign.
AI is already inside the payment flow
That legal assumption is colliding with operational reality. AI systems are already making consequential decisions upstream of payment execution. A human sets parameters on Monday. The agent acts inside those parameters at 2 a.m. on Thursday. The transaction still reaches a settlement layer that was built to recognize credentials, not delegated machine authority.
The IMF called out this exact tension in its April 22 note, How Agentic AI Will Reshape Payments. Their framework breaks agentic payments into three layers: intent, authorization, and settlement. The important point sits in the middle. Intent can still start with a human. Settlement can still run on established rails. Authorization is where the model breaks.
Current market solutions treat that gap as a controls problem around the edges. Single-use credentials, pre-set limits, narrow delegation windows. Those tools help. They do not solve the core issue. They proxy for live authorization. They do not establish what the settlement layer should recognize once an agent starts acting on its own timing and within its own decision path.
Why this becomes a liability problem
That proxy works at small transaction values. It gets fragile at enterprise scale, where agents negotiate terms in real time, route spend dynamically, or hand work to other agents. Once delegation chains start moving across systems, the industry loses the clean handoff that made old authorization rules workable.
The problem is not that existing frameworks reject agentic payments. The problem is that they are structurally silent on them. In payments infrastructure, silence does not stay neutral for long. It turns into liability disputes, fraud claims, inconsistent enforcement, and expensive arguments over who actually authorized the movement of funds.
That is why this issue matters now. AI in commerce is arriving faster than the settlement layer can reinterpret its own assumptions. If the execution layer scales before the authorization layer catches up, the market will learn the hard way, one disputed transfer at a time.
What needs to happen next
The industry needs an agent attestation standard. Not a new law, a new authorization primitive that sits beside identity verification in the settlement stack. Who set the parameters? When did they set them? What constraints were in force? Did the agent stay inside that scope? Did another agent inherit the authority? Those are answerable questions if the infrastructure is designed to preserve the chain.
The institutions best positioned to build that standard are the ones that already govern settlement trust. The Fed. NACHA. Card networks. Banks that sit on standards bodies. They are not the fastest actors in the market. They are the actors whose recognition matters when money actually moves.
The Florentine banker's seal verified authority, not identity. Modern payment infrastructure needs to recover that distinction. The system does not need to prove the agent is human. It needs to prove the agent is acting with valid authority. That is the real gap opening inside agentic commerce.