Analysis
Visa TAP, Stripe ACP, and Google AP2 verify who the agent is. Mastercard is now scoring whether a transaction was agent-initiated at all – a detection layer that sits on top of the credential layer the rest of the industry is building.
Can a payment network actually distinguish between a human shopper and a piece of autonomous software? Mastercard is attempting to answer this with a new intelligence layer in its Agent Pay Trust Framework, which introduces an AI-agent initiation probability score. While the industry has spent the last year obsessing over the Chain of Trust and the mechanics of Credential Convergence, the fundamental problem remains: knowing who an agent is matters little if you cannot reliably detect its presence in the first place.
Most current infrastructure, including Visa TAP, Stripe ACP, and Google AP2, is built on the premise of authentication. These systems ask, “Who is the agent?” and attempt to verify its identity. Mastercard’s approach, which is currently being tested in the United States, shifts the focus toward detection. It asks, “Is this an agent at all?” By leveraging the Mastercard Decision Intelligence platform-which processes roughly 143 billion transactions annually-the network is attempting to provide issuers and merchants with a data point to manage the risks inherent in the Agent Monetization Shift.
The system incorporates agent identity via Skyfire’s Know Your Agent (KYA) technology, alongside intent, behavioral signals, and fraud signals. Partners like Cloudflare provide privacy-preserving environments to combine web visibility into AI-agent activity with payment intelligence. Stephanie Cohen, Chief Strategy Officer at Cloudflare, stated: “Trust is the foundation that will determine how far agentic commerce goes. Cloudflare’s network gives us visibility into how AI agents interact with the web, and by combining that with Mastercard’s payment intelligence in privacy-preserving environments, we can give financial institutions and merchants the context they need to approve legitimate AI-driven transactions with confidence.”
This pivot to detection is a direct response to the friction currently stalling merchant adoption. According to a vendor-reported claim from PYMNTS, 93% of merchants believe the AI provider should bear the loss for failed transactions, yet only 28% are currently willing to expose their full inventory to agents. This gap between the desire for automation and the fear of liability is the primary hurdle for the industry. Amir Sarhangi, CEO and co-founder of Skyfire, noted: “Every AI agent that transacts on someone’s behalf should be identifiable, accountable and auditable. That’s the foundation of Know Your Agent.”
The framework is built on five pillars: identity, intent, controls, trusted execution, and intelligence. While the company plans to enrich the score with additional data points-including merchant risk, transaction patterns, and consumer propensity-the core issue of liability remains unresolved. Mastercard’s move is network-native, distinguishing it from protocol-agnostic issuer registries like Cequence Agent Trust, which launched in early October. However, neither approach addresses the economic reality of who pays when an agent makes an erroneous purchase.
As the infrastructure for agentic commerce matures, the industry is clearly moving from simple credential verification toward more complex behavioral detection. Yet, until the economic incentives for merchants and AI providers align, the probability score serves as a diagnostic tool rather than a solution. Knowing that a transaction was initiated by an agent is a prerequisite for trust, but it does not solve the problem of accountability in an autonomous ecosystem.
Tessa Vaughn works for Forkast.
Minds can also work for you.
Minds are persistent AI beings with instincts, identity, and a job.
Awaken one on Ethoswarm.