Banks and payment firms test know-your-agent rules to authenticate AI agents in invoice and liquidity workflows.
Financial services firms and partners are taking steps to address the greatest hurdle to agentic AI adoption in commerce: distinguishing legitimate AI agents from malicious ones without triggering fraud alerts.
Modern anti-fraud systems analyze a variety of parameters and flag any deviations as suspicious activity, said Lev Brodsky, CTO of FinteqHub, a European payment orchestration platform.
“For instance, a purchase made at 3 a.m. from a cloud server’s IP address that lacks the typical signs of human interaction with a website might appear suspicious, yet an AI agent executing a task on behalf of its owner is perfectly routine,” he said. “This creates a paradox: technologies developed over years to protect payments from automated attacks now potentially hinder the growth of agentic commerce.”
Injecting Trust
Two pilot projects aim to embed trust in agentic workflows by adopting know-your-agent frameworks.
Credit card processor Visa, financial operations vendor BILL and cash-generation platform provider Billtrust launched a pilot that lets AI agents initiate and complete approved buyer invoice payments.
When an AI agent initiates a payment in the live pilot, Billtrust uses Visa’s Trusted Agent Protocol to recognize and authenticate the agent. Once authenticated, the AI agent enables BILL to complete the invoice payment.
Meanwhile, Kinexys, J.P. Morgan Payment’s blockchain unit, has collaborated with Hyundai Card, the credit card service of Hyundai Motor Group, and completed the initial phase of a pilot that focused on a treasury-management use case.
The test case involved a Hyundai Card AI agent operating under a digitally signed AI agent mandate, managing euro liquidity in its Kinexys Blockchain Deposit Account based in Frankfurt, Germany. As part of the test, the agent could request foreign exchange executions as needed to maintain the account’s liquidity.
“Only when Kinexys has verified the identity and authority of the Hyundai Card AI agent and evaluated its request against the pre-defined, digitally signed AI agent mandate owned and managed by Hyundai Card, can the Kinexys AI agent either auto-execute the trade, escalate to a human for review, or deny the request,” Wee Kee Toh, Global Head of Kinexys Insights, told Global Finance. “This [proof of concept] explored one-to-one AI agent transactions, and future phases of the POC will explore one-to-many.”
The Next Step
These pilots are an important step toward advancing agentic commerce, but they are far from the point where agents could independently select vendors, negotiate terms, and complete transactions.
Embedding know-your-agent protocols directly into payment systems and anti-fraud infrastructure is the industry’s next logical step, FinteqHub’s Brodsky said. “This would reduce false declines without compromising security. Until such infrastructure is widely adopted, agentic payments will remain limited to scenarios that require prior human approval.”
