A Layered Cryptographic Trust Architecture (LCTA) for Secure Agentic Commerce in Autonomous AI Marketplaces

Anath Bandhu Chatterjee

Payment Systems Architecture

San Jose, California, USA

Abstract—The emergence of agentic commerce — autonomous AI agents discovering, negotiating, and settling purchases on behalf of users — has progressed from prototype to production within twelve months. Visa's Trusted Agent Protocol (TAP), Google's Agent Payments Protocol (AP2), Stripe / OpenAI's Agentic Commerce Protocol (ACP), Coinbase's x402, and Cloudflare's Web Bot Auth now deliver fragments of a trust fabric, yet no single specification spans the full lifecycle from user intent to settled payment with non-repudiable evidence. This paper presents LCTA, a four-layer cryptographic trust architecture that composes existing primitives into a coherent end-to-end framework. Layer 1 (Cryptographic Agent Identity) anchors every agent to a Decentralized Identifier (DID) and signs every HTTP request via RFC 9421 Message Signatures. Layer 2 (Verifiable Intent Binding) introduces the Intent-Bound Transaction Token (IBTT), a hash-chained construction that fuses Intent, Cart, and Payment Mandates into a single non-malleable artifact whose modification voids the transaction. Layer 3 (Risk-Adaptive Authorization) computes a composite risk score R(t) over five orthogonal signal classes and emits an allow / step-up / deny decision aligned with PSD3 Strong Customer Authentication and EMV 3-DS v2.3. Layer 4 (Continuous Verification) records every decision to an append-only attestation log, supplying chargeback and dispute-grade evidence. We present a formal threat model spanning the OWASP Agentic AI top-15 categories, indirect prompt injection (IDPI), and replay / confused-deputy attacks; demonstrate a 25.0 / 26.0 threat-coverage score versus 17.0 for the strongest prior single-protocol baseline; and discuss compliance alignment with PCI-DSS v4.0.1, GDPR Article 22, and the EU AI Act. The paper is intended as an actionable design reference for payment networks, issuers, and merchants integrating agent-initiated commerce in 2026 and beyond.

Index Terms—Agentic commerce, AI security, payment systems, verifiable credentials, decentralized identity, trust architecture, AP2, Trusted Agent Protocol, risk-based authorization, PSD3, EMV 3-DS, intent binding.