Thibault Murel

AI agents are already initiating real financial transactions - booking, subscribing, settling, and disbursing without a human approving each step. The commerce and payments infrastructure built around human decision-makers was not designed for a counterparty that never sleeps, cannot be held to a contract in the traditional sense, and can be compromised at the model layer rather than the credential layer.

This session convenes payments architects, legal practitioners, and AI deployment leads to work through the practical questions the industry is looking to answer: how do you authenticate an agent, assign liability for an autonomous transaction gone wrong, and design governance frameworks for a technology that is already in production.

Participants will examine the specific mechanics of agentic payment flows - wallet delegation architectures, spending limit enforcement, multi-agent authorisation chains, and the emerging standards around agent identity that are still being written in real time by consortia that have not yet reached consensus.

The session will also surface the fraud and risk management implications: how transaction monitoring systems designed to flag anomalous human behaviour respond to the high-frequency, pattern-consistent activity of a well-functioning agent - and what entirely new risk frameworks may be required when the buyer is a model, the merchant is an API, and the dispute resolution process was designed for neither.