Proof that a human stands behind an agent becomes a banking product

This space is crowded. Proving a human stands behind an agent is already sold by World AgentKit, Skyfire KYA, Persona and Zerohash. Recording delegated spend limits is built into Mastercard Agent Pay tokens, Visa's Trusted Agent work and Google's AP2 mandates. So a standalone Delegation Registry for issuers and processors would be competing directly with the card networks on their own rails, and that is a weak spot for a startup. The more open opening is on the consuming side of One Human, N Agents, aimed at contact centers. Nearly all the current credentials target checkout and web traffic. None of the products I found turn them into rate limits, dedupe or policy keyed to the person for non-payment endpoints like complaints, refunds, disputes, chargebacks and account servicing, or for voice and chat channels where agents increasingly call in. A neutral layer could take in World ID, Skyfire, Persona, network tokens and AP2 mandates, resolve them to one stable person-level identifier for each bank, and enforce per-human quotas and refund-abuse rules in a fraud team's existing stack. That layer could still win, especially if it is sold through or alongside contact-center platforms and Pindrop-style fraud tools rather than against them. The risks are that Pindrop or Persona add this as a feature, and that World or Skyfire ship their own policy dashboards. Speed and owning the integrations with CCaaS vendors would matter more than the credential itself.

• World (Tools for Humanity) AgentKit

  Beta SDK launched March 2026. Lets a site check that a unique, iris-verified World ID human stands behind an agent. It allows several agents per human and uses zero-knowledge proofs, and it is wired into x402 payments with Coinbase. This is almost exactly the One Human, N Agents primitive. Its weak point is that it depends on Orb biometrics.

• Skyfire KYA / KYAPay

  Issues JWT credentials that name the agent, the human or enterprise it represents, and whether it was verified. Agents inherit their owner's verification. Bot-management vendors like DataDome and Fastly check the tokens at the edge, and Skyfire has demoed this with Visa Intelligent Commerce and Trusted Agent Protocol.

• Persona Know Your Agent

  Sells KYA tooling built on its KYC/KYB stack to verify the people and businesses behind agents. Lithic, Skyfire, Crossmint and OpenAI are cited as customers.

• Mastercard Agent Pay (Agentic Tokens)

  Network tokens tied to a specific agent and a consumer consent policy, with spend ceilings, merchant categories and time windows set when the token is issued. This already covers much of the Delegation Registry idea, inside the card network where issuers and processors can read it.

• Visa Trusted Agent Protocol / Intelligent Commerce

  Visa's protocol lets merchants recognize trusted agents and accept authenticated intent from them on a consumer's behalf. It is the competing network-level delegation and trust layer.

• Google AP2 (Agent Payments Protocol)

  Defines Intent, Cart and Payment mandates as signed verifiable credentials that record what a user authorized an agent to buy. It is an open standard for the delegation-record concept. This comes from a secondary source.

• Zerohash Agentic Finance Suite

  Reported July 2026. Requires a screened person to be recorded as accountable for any agent that spends, with session spending caps set in advance. This is close to a delegation registry on the infrastructure side.

• Pindrop

  Incumbent in contact-center fraud: voice and device risk scoring, deepfake detection embedded in Zoom Contact Center, and refund-abuse use cases. It is not yet tied to agent-to-person credentials, but it owns the buyer this idea would target.

Checked 2026-10-19