Visa, Mastercard Launch Financial Automation for AI Agents

In March, Mastercard completed its first autonomous agent payment in Hong Kong. An AI agent booked a ride-share, authenticated through HSBC, and paid without a human touching the screen. The same week, Visa launched its Trusted Agent Protocol partnership with Ramp, rolling out automated bill payments across tens of thousands of corporate clients. Two separate card networks, two different technical approaches, both shipping production infrastructure for a transaction channel that did not exist a year ago.

These are not chatbot integrations or AI-powered search features layered onto existing checkout flows. Visa and Mastercard are building systems where AI agents independently identify purchases, negotiate terms, authenticate themselves, and settle payments from start to finish. The scale and speed of the rollout reflects how seriously both companies view what comes next for financial automation across banking, insurance, and wealth management. This is banking automation at a structural level — new transaction architecture built from the ground up for agents.

What Visa and Mastercard Actually Built

Visa's program, called Intelligent Commerce, centers on a Trusted Agent Protocol designed to solve the identity problem that has stalled autonomous agent payments. When an AI agent tries to buy something on behalf of a consumer or business, the merchant needs to know the agent is authorized to spend. The Trusted Agent Protocol establishes that chain of trust: validating the agent's identity, confirming its authorization scope, and securing the transaction data throughout the process. Visa built this in partnership with Akamai and Amazon Web Services, and paired it with AI-driven dispute resolution tools that already handle a significant share of global payment disputes each year.

"I haven't seen anything like this since the dawn of ecommerce itself in the late '90s or early 2000s," said Jack Forestell, Visa's chief product and strategy officer. "The agent needs an identity. You need to secure that identity, you need to validate it, you need to collect more data to ensure the security."

Mastercard's equivalent is Agent Pay, paired with a trust layer called Verifiable Intent. Where Visa focused first on corporate payments through the Ramp partnership, Mastercard moved into consumer transactions. Its Hong Kong pilot with HSBC processed real payments for ride-share services, with the agent handling the full transaction lifecycle autonomously. Mastercard has since expanded Agent Pay to Australia, the United States, and India, with early banking partnerships at Citi and US Bank.

Both companies integrated with Stripe's Shared Payment Tokens, which let agents reference tokenized payment credentials without seeing raw card data. Google's Universal Commerce Protocol, announced in January, provides another interoperability layer. The fintech AI ecosystem forming around agent payments is growing fast. Affirm, Klarna, Cloudflare, Shopify, Checkout.com, and Adyen have all announced adjacent capabilities this year.

The identity and authorization challenge is where the real engineering happens. An AI agent paying for a SaaS subscription on behalf of a procurement team requires cryptographic proof of authorization, scoped spending limits, and audit trails. Traditional payment flows were never designed to provide any of that. Visa's Trusted Agent Protocol and Mastercard's Verifiable Intent layer are purpose-built to handle authentication for an entity that is not a human and does not have a physical card.

AgentPMT's x402Direct protocol approaches the same authorization problem from outside the card networks entirely, using smart contracts on Layer-2 chains to enforce spend caps and authorization boundaries for autonomous stablecoin payments. The card network approach and the on-chain approach represent two different engineering answers to the same fundamental question: how do you let an agent spend money without giving it unrestricted access to the underlying account?

The Market Already Moved

The urgency behind these launches makes more sense against the spending data. According to Salesforce, AI agents influenced $262 billion in U.S. holiday sales during the 2025 season, roughly one-fifth of all retail transactions. That figure captures purchases where an AI system recommended, compared, or completed part of the buying process. It is not a forecast or a projection. It already happened.

The behavioral shift in financial services was just as sharp. A TD Bank study found that 55 percent of Americans consulted large language models for financial advice in 2025, up from 10 percent the year before. That kind of adoption curve does not happen gradually. It means millions of consumers are already using AI tools to evaluate lending terms, compare insurance quotes, and weigh investment options before making purchasing decisions.

Robinhood launched an AI advisory product that attracted a significant paying subscriber base within months. Anthropic partnered with LPL Financial to bring AI-driven guidance to a large network of advised clients. The financial advisory industry is watching a generational retirement wave approach at the same time AI tools are proving they can handle routine portfolio analysis and product comparison. The demand for agent-capable financial infrastructure is being pulled forward by consumer behavior that has already changed.

For banks, insurers, and wealth managers, the implication is direct. If an AI agent recommends financial products to a consumer, the agent needs to read your product terms, compare your rates, and process a transaction. Firms whose products are not machine-readable will not surface in agent-driven recommendations. As Yaacov Martin, CEO of lending platform Jifiti, wrote: if an agent cannot read your credit product, you do not exist in that context.

AgentPMT's Dynamic MCP addresses this visibility problem from the vendor side. Instead of requiring financial products to be individually integrated with each AI platform, Dynamic MCP lets vendors list tools and services in a marketplace that agents can search and access on demand, loading only what they need without consuming the agent's working memory with unused product definitions. For financial services firms trying to become agent-readable, a marketplace-based approach removes the need to negotiate separate integrations with every AI platform individually.

Three Regulatory Frameworks, Six Months

The regulatory picture is catching up to the technology, but it is arriving in pieces, and the deadlines are tight.

The National Association of Insurance Commissioners launched a 12-state pilot in March 2026, examining how insurers use AI in claims decisions, total-loss valuations, and damage assessments — the core functions where AI insurance claims processing has expanded fastest. California, Colorado, Florida, and nine other states are participating in the evaluation, which runs through September before a potential nationwide rollout vote in November. The insurance industry formally objected, arguing the program is voluntary for regulators while compulsory for companies. The NAIC responded that existing state insurance laws apply regardless of whether decisions come from humans, algorithms, or third-party vendors. Insurers that rely on third-party AI platforms for claims processing are on notice: the regulator considers the insurer responsible for the AI's output, regardless of who built it.

Colorado's AI Act, the first state law specifically targeting high-risk AI systems in financial services, takes effect June 30, 2026. It covers AI used in lending decisions, credit scoring, and fraud detection, requiring documented decision pathways and algorithmic impact assessments for any system that materially affects consumer outcomes. The EU AI Act follows in August, requiring compliance from any high-risk financial AI system operating in or serving European markets. Between state law, federal pilot programs, and international regulation, financial services firms face three distinct compliance frameworks converging within six months.

The adoption of agentic AI in banking has outpaced the governance structures meant to oversee it. An EY analysis found that a large majority of banking firms now report using some form of agentic AI, but most lack the compliance infrastructure to satisfy what Colorado, the EU, and the NAIC are preparing to enforce. Firms need audit trails for agent decisions, documented governance structures mapped to frameworks like NIST AI RMF or ISO 42001, and clear accountability chains for any third-party AI systems they integrate. AI compliance tools that generate these audit trails automatically are moving from optional to essential.

AgentPMT's budget controls and human-in-the-loop mobile approval system produce the kind of auditable decision trail that these regulatory frameworks increasingly require. Every agent action runs through scoped permissions, spending limits, and credential access controls, with human approval required for sensitive transactions. For firms evaluating how to meet overlapping compliance deadlines, built-in governance reduces the distance between current operations and regulatory expectations.

What Financial Services Firms Should Do Now

The firms that will capture distribution in an agent-driven market are the ones making their products accessible to AI systems now, not after the infrastructure is settled.

That starts with structured, machine-readable product data. APIs, Schema.org formatting, and well-defined metadata allow an AI agent to parse lending terms, insurance premiums, and portfolio options without human interpretation. For lenders, that means publishing rate structures and eligibility criteria in formats that agents can query directly. For insurers, it means structured policy data that an agent can compare across carriers without scraping marketing pages. Insurance automation that stops at internal workflow efficiency misses the distribution shift — the value is in making products findable by agents already advising consumers.

It extends to governance frameworks that satisfy regulatory requirements across jurisdictions simultaneously, since Colorado, the EU, and the NAIC are all operating on different timelines with overlapping scope. And it includes payment infrastructure that works with agent-driven transaction flows, whether those run through card networks, stablecoin rails, or both.

McKinsey projects that AI agents could drive $1 trillion in U.S. transactions by 2030. The infrastructure to handle that volume is being built this quarter, by Visa, Mastercard, Stripe, and a growing ecosystem of companies engineering the payment and identity layers for autonomous commerce. The financial services firms that make their products visible to agents this year will be positioned when that transaction volume arrives. The ones that wait will find themselves competing for distribution in a channel they never built for.

---

Sources

• US Card Networks Accelerate Bets on Agentic AI — American Banker
• How Visa and Mastercard are Approaching Agentic Commerce — Digital Commerce 360
• The New Invisible Battlefield for Banks: AI Drives $262 Billion in Sales — FinTech Weekly
• Should You Trust AI to Manage Your Money? — Fortune
• Regulators Open First Examination of Insurer AI — Autobody News
• Four Regulatory Shifts Financial Firms Must Watch in 2026 — EY
• Colorado's Landmark AI Law Coming Online — Brownstein