Visa and OpenAI partner to secure agentic commerce AI purchases

Agentic commerce, in which AI acts as an automated buyer for consumers and businesses, is rapidly emerging as a key technology trend, exemplified by Google’s recent Universal Cart launch. This week, Visa and OpenAI announced a significant partnership on Wednesday to embed payment systems into AI platforms, aiming to enable secure, agent-led transactions.

The collaboration introduces Visa’s Trusted Agent Protocol to OpenAI interfaces like Atlas and ChatGPT Shopping. This system allows developers and retailers to process payments confidently initiated by autonomous AI agents.

Visa states that the system operates within strict, user-defined boundaries. Consumers and businesses can set specific spending caps, merchant categories, and approval requirements to keep full financial control while allowing AI to handle checkout.

For ordinary users, this means entrusting an AI with personalized shopping research and routine purchases. For merchants, it offers a seamless buying experience across the expanding digital platforms where consumers spend their time.

Industry-wide, the effort to secure AI shopping is gaining momentum. Following earlier advances such as Google’s Agent Payments Protocol, Mastercard introduced Agent Pay for Machines, a competing system for high-frequency, low-latency transactions between AI entities.

Despite rapid progress, entrusting AI with financial power raises trust issues. Cybersecurity experts express concerns about rogue agents, unauthorized purchases, or AI tools leading users to fraudulent websites.

In response, Visa highlighted that transparency is central to their approach, with user permissions, tokenized credentials, and real-time fraud detection actively in place to prevent unauthorized transactions.

Nevertheless, Geoff Cairns of Forrester warned that agentic payments pose unique challenges. While existing security measures are strong, the main risk now lies in ensuring that the autonomous agent acts in accordance with the buyer’s true intent and policies.

Cairns pointed out that issues like liability, mistaken transactions, and the potential for fraud are major concerns as this technology advances.