Antom, the merchant payment and digitization services provider under Ant International, has introduced an agentic payment solution designed to support AI-driven transactions.
The solution includes a secure alternative payment method (APM) checkout system, which Antom said makes it one of the first partners of Mastercard and Visa to pilot card-based transaction features for AI agents.
Antom said the new offering provides an AI-ready payment mandate model and advanced asset management to accurately identify user intent while strengthening security and improving transparency. It also supports multiple payment options with enhanced risk management tailored for AI agents.
“Agentic payment is a foundational step in allowing AI agents to generate real value in our everyday life,” said Gary Liu, GM of Antom, Ant International. “The rise of agentic payment calls for rethinking how payment systems are designed. We look forward to co-building the protocols and frameworks with partners across the financial, tech, and commerce sectors to ensure agentic payments are smooth and reliable.”
The solution builds on the Model Context Protocol (MCP) and enables embedded payment flows through dialogue-based interactions with AI agents. This includes confirmed purchase requests and pre-authorized transactions such as scheduled sales or spending within a set limit. Antom has made the system open-source on GitHub.
Unlike traditional checkout systems that require multiple clicks, agentic payment confirms intent during natural language conversations. To support this, Antom connects AI agents to a wide range of digital wallets and ensures compliance with PCI Security Standards Council (PCI SSC) requirements.
The system uses cryptographic measures to manage payment assets and applies intent analysis models for each transaction. By linking transaction details with intent evidence, it generates verifiable credentials that create end-to-end traceability. This allows users to keep visibility over how AI agents act, with disputes resolved through privacy-computing-based credential queries.