Know Your Agent: Verifying the Customer Who Isn’t a Person
AI agents are entering the financial ecosystem, both as analytical assistants for finance professionals and as tools that move and manage money for businesses on their own. On Slash, businesses can already connect an agent to create cards, send payments, and manage invoices.¹ AI is taking on tasks people used to handle, and our response so far has been to assume the old ways of doing things will keep working. They won’t.
An important concept in the financial industry is due diligence: the requirement that a bank or financial provider verify the identity of the individual or business it serves. The goal is to make sure a bad actor isn’t posing as someone else to abuse the provider’s services, and to have proper identity records on file if recourse or legal action is ever needed.
If you have a bank account or credit card in your name, you’ve been through due diligence before. You gave the bank your Social Security number and personal information, maybe submitted financial statements and history, and perhaps met with a banker in person to finish an account review. The process is built on documents that prove a person exists and a relationship between the bank and its customer.
An AI agent can’t fulfill those requirements.
The financial industry has moved quickly to create a new type of due diligence to address the problem: Know Your Agent (KYA). The name follows the pattern of existing due diligence, Know Your Customer (KYC) and Know Your Business (KYB). As agentic finance becomes a more integral part of managing money, more services will start asking about the what, who, where, and why of how you use an agent. This guide covers where those questions are coming from and how your business can answer them.
The standard in finance
Slash goes above with better controls, better rewards, and better support for your business.
Why Verifying an Agent Is Harder Than Verifying a Person
To understand the challenge of knowing an agent, it helps to start with how KYC deters fraud in the first place.
When a bank asks for your name, date of birth, Social Security number, and address, it’s collecting what it needs to confirm you are who you say you are. It checks that information against identity databases, government sanctions lists, and records from other institutions, so serving you won’t make it complicit in fraud, identity theft, money laundering, or other financial crimes.
Verifying an agent starts with verifying the person behind it because identity verification has nothing to anchor to without a human. From there, the process breaks into three steps. Onboarding confirms the person exists. Verification confirms the human using the agent is who they say they are. Ongoing authorization confirms the agent is permitted to act on that human’s behalf.
Due diligence doesn’t end once your account is open. Financial providers are required by law to keep records of your activity, which builds a profile of how you use their services. The profile powers fraud detection: if you’ve ever received an alert about an unusual transaction, it came from those records.
Agents complicate the profile. They run around the clock and transact in patterns that don’t look like a person’s, which muddies what counts as normal activity for the account holder.
Even if a bank trusts you as the account holder, how can it trust the agent acting on your behalf every day after?
How the Financial Industry is Approaching KYA
Visa and Mastercard are leading the push for KYA with their own frameworks, and Google built the protocol both card networks now operate around. Here’s how each works:
Visa Trusted Agent Protocol
Visa unveiled Trusted Agent Protocol in October 2025, developed in collaboration with Cloudflare. The problem it targets sits on the merchant’s side of checkout. Bot detection systems can mistakenly block legitimate agent transactions, and merchants risk losing visibility into the consumer behind the agent.
TAP gives approved agents a cryptographic signature to attach to their requests. The signature carries the agent’s intent to view or buy a specific product, signals showing whether the consumer already has an account with the merchant, and optionally the payment data needed to check out. Each request is locked to the merchant’s specific website and the exact page the agent is on, so bad actors can’t capture an old request and reuse it.
The approval matters as much as the signature. Under Visa’s specification, agent providers are AI platforms that have been onboarded to the payment network and approved to make payments on behalf of account holders. A key associated with Visa signals the agent complies with Visa Intelligent Commerce program requirements. In KYC terms, the network handles onboarding and the merchant runs the check at the door.
TAP has since become one piece of a larger program. In April 2026, Visa launched Intelligent Commerce Connect, a single integration that lets merchants accept agent payments across TAP and other major protocols, with both Visa and non-Visa cards.
Mastercard Verifiable Intent
Visa’s protocol tells a merchant whether an agent is legitimate. Mastercard’s tells everyone in the transaction what the human allowed the agent to do.
Mastercard introduced Verifiable Intent in March 2026, co-developed with Google, as the newest layer of its Agent Pay program. Mastercard’s argument starts with the checkout button. When you tap a card or click “buy,” every party understands what you chose. With agents, consumers delegate instead, and intent becomes less visible at the moment of the transaction even as it becomes more important to verify.
Verifiable Intent links identity, intent, and action into one privacy-preserving record. It confirms the cardholder who authorized the agent, captures the consumer’s specific instructions, and records the interaction between agent and merchant that produced the purchase. If a dispute comes up, all parties work from the same audit trail, and selective disclosure means each party sees only the minimum information it needs.
The limits are the point. Mastercard describes agentic tokens that can be restricted by agent, merchant, category, spending limit, timeframe, or usage rules, with verifiable intent tying the user’s permission to the resulting transaction. The trust problem is real on the consumer side: Mastercard’s own research found only 10% of consumers are willing to let an agent complete a purchase autonomously.
Google Agent Payments Protocol
Google launched AP2 in September 2025 as an open, payment-agnostic protocol for agent purchases, including ones where the user isn’t present. Where Visa checks the agent and Mastercard records the authorization, AP2 handles the record of the whole purchase.
Every agent transaction carries three signed mandates. The intent mandate records what the user asked for, the cart mandate records what the agent assembled, and the payment mandate tells the network an agent is involved and what will be charged. Together they prove what the user authorized and define the limits of the agent’s purchasing power.
The intent mandate is where delegation happens. A user can pre-sign it with rules like price limits and timing, and the agent builds the cart on its own once those conditions are met. AP2’s mandates are the closest thing yet to a formal standing instruction for an agent, and the chain from intent to payment leaves an audit trail for disputes.
Google has since handed AP2 to the FIDO Alliance to keep it platform-agnostic, and Mastercard designed Verifiable Intent to work alongside it. Google’s roadmap also extends AP2 to real-time payment systems like India’s UPI and Brazil’s Pix.
Other KYA efforts worth knowing about
- OpenAI and Stripe’s Agentic Commerce Protocol (ACP) is a checkout protocol more than a KYA framework, but its Shared Payment Token is a clean example of scoped authority. The token lets an app like ChatGPT pay without seeing the buyer’s card, and it’s single-use, time-bound, and capped at a set amount.
- Skyfire is one of the few companies to brand its product as KYA outright. Its KYAPay protocol issues a signed token carrying verified information about the agent’s owner, and Skyfire has demonstrated purchases running KYAPay alongside Visa’s TAP.
- Sumsub launched a KYA module for MCP in October 2026. It logs agent activity, enforces financial thresholds, and triggers a liveness check when an agent attempts a high-value action, confirming a verified human approved it.
- NIST published a concept paper on AI agent identity and authorization in February 2026, covering identification, authorization, auditing, and non-repudiation. It’s early, but it signals KYA may move from industry practice toward government guidance.
Interoperability: How Different KYA Frameworks Fit Together
KYA is complicated, and no one is handling all of it alone. Visa is focused on the merchant’s verification, Mastercard on leaving a trail of intent, and Google on the purchase from first instruction to final payment. Covering the full scope of KYA requires those efforts to fit together into one interoperable system.
The FIDO Alliance, the standards body behind secure passkeys, formed an Agentic Authentication Technical Working Group in April 2026, focused on how users can delegate actions to AI agents securely. A parallel Payments Technical Working Group is developing specifications for agent-initiated commerce. Both draw from Google’s AP2 and Mastercard’s Verifiable Intent, and the payments group is chaired by members from Mastercard and Visa.
However, a settled standard for the payments industry is still a way off. According to an October 2026 analysis from Boston Identity, FIDO hasn’t given a date for turning the two contributions into a specification. In the meantime, payment companies are supporting several protocols at once. Visa’s Intelligent Commerce Connect, for example, lets merchants accept agent payments from four competing protocols through a single integration.
How to Prepare Your Business for KYA
McKinsey estimates AI agents could orchestrate $3 trillion to $5 trillion in global consumer commerce by 2030. That figure only covers goods, meaning it leaves out services and the B2B market entirely. Business use is already underway: Visa’s early Intelligent Commerce pilots included B2B purchases and corporate bill pay.
Moving money with AI is becoming normal, and the systems that make it workable are maturing. Before long, the financial providers you work with will start asking the KYA questions. Here’s how to have answers ready.
What agent is acting?
Start with an inventory of every agent that can interact with your money: the platform it runs on, the tasks it handles, and the accounts it can reach. Find out whether each agent’s provider participates in a network program like Visa Intelligent Commerce or Mastercard Agent Pay, since those programs are where agents get approved in the first place.
Give each agent its own identity. NIST’s concept paper frames agents as entities that should be identifiable on their own, not anonymous automation running on shared logins. An agent using an employee’s credentials is indistinguishable from that employee, which makes its activity impossible to audit.
Who’s connected to the agent’s action?
Every framework above starts with a verified human or business. For most companies, KYA is built on KYB: the entity records, beneficial ownership details, and authorized signers you already provided to open your accounts. Keep them current.
Then name an owner for each agent. When a provider asks who authorized an agent to act, the answer should be a specific person with the authority to approve the agent’s spending, not “the finance team.”
Where can the agent spend?
The networks are building agent permissions around limits: merchant, category, amount, and timeframe. Set the same limits yourself before anyone asks. An agent paying software subscriptions doesn’t need access to payroll, and an agent booking travel doesn’t need an unlimited card. The cleanest setup is a dedicated card or account per agent, with spending rules attached to it.
Why did the agent act?
Verifiable Intent and AP2 both exist to answer one question after the fact: did the human authorize the payment the agent made? Your own records should answer it too. Keep a log of the instructions each agent receives, the approvals it gets, and the transactions that result.
Route large or unusual payments through a human approval step, and reconcile agent activity against your books on a regular schedule. When a dispute arrives, the log becomes your evidence.
How Slash is Building for Agentic Finance
Networks and merchants will run the KYA checks, but the answers come from your own accounts: what an agent could reach, what it was allowed to spend, and what it left behind. Slash keeps all of it in one place:
Orchestration. An agent making a payment has to choose how to pay, clear an approval, send the money, and record it. Orchestration refers to how those steps are ordered and controlled within one system, and the Slash MCP is how an agent connects to it. Slash handles orchestration across cards, bank transfers, and stablecoins, providing a clear schema of available actions, so any agent can understand what it can do and how to do it.⁴ Card numbers are encrypted before they reach the agent, so it can pay without ever seeing them.
Auditability. Activity Logs record every action on an account, who took it, and what changed, down to the before-and-after values when someone edits a rule or card limit. AI-generated categorizations are marked, and transactions sync to your accounting software.
Financial controls. Agents can draft and submit payments, but nothing executes until someone on your team signs off. Multi-step approvals can route a payment through more than one reviewer before it goes out. Card groupings let you set rules by team for how and where money can be spent, and the same rules can scope what an agent is allowed to do. Real-time fraud alerts from Slash learn from your transaction patterns over time and flag activity that falls outside them.
Providers will soon ask what your agent is, who authorized it, where it can spend, and why it paid. With Slash, you’ll have the answers before they ask. Get started with a Slash account today with the form below.
Apply in less than 10 minutes today
Join the 10,000+ businesses already using Slash.
