American Express is finally acknowledging the elephant in the room: you cannot build a commerce network on top of autonomous agents if nobody trusts the agents to actually pay for anything. On October 6, 2026, the company released its Amex Business Playbook for Agentic Commerce, a document that effectively admits that their earlier technical work – specifically the ACE Developer Kit launched in April 2026 – was just the foundation. Now, they are trying to build the guardrails.
The industry has been stuck in a holding pattern where the technical plumbing exists, but the liability and trust models are essentially blank pages. Anna Marrs, Group President of Global Merchant & Network Services, is now trying to steer the firm into this vacuum. The playbook introduces two main levers: the Agent Purchase Protection initiative and a pilot Merchant AI Advisory Council. It is a classic move for a legacy player: when the market gets messy, create a committee and promise to pay for the mistakes.
The Agent Purchase Protection initiative is the headline grabber, though it is important to keep our feet on the ground. As of now, this is an announced intent, not a live, functional product. AmEx says it wants to protect eligible Card Members and merchants from charges caused by errors from registered AI agents. But the devil is entirely in the details of the terms and conditions, which are currently missing. By positioning this as an industry-first, AmEx is trying to solve the liability gap. It is a smart play, especially considering that 93% of merchants currently believe AI providers should be the ones footing the bill for agent errors. AmEx is essentially volunteering to step into that liability vacuum, provided they can define what a ‘registered’ agent actually looks like.
The urgency here is backed by their own Trendex survey data, which highlights a massive disconnect. While 83% of businesses are salivating over the opportunity to use AI agents, consumers are not buying it. Only 15% of shoppers trust bots to handle big-ticket purchases, even though 66% are happy to use them for simple price comparisons. This delta is the real constraint on agentic commerce. If the consumer doesn’t trust the agent to get the transaction right, the business case is just a spreadsheet exercise.
To bridge this, AmEx is forming a Merchant AI Advisory Council. The goal is to bring merchant perspectives into the room as they figure out how these agents interact with the network. We have already seen how retailers are split between welcome, browse, and block postures. A council is a standard way to try and harmonize these conflicting approaches into a single, coherent network standard. It is a necessary step, but it remains to be seen if merchants will actually get a seat at the table or if this will just be a feedback loop for AmEx’s own product roadmap.
This initiative doesn’t exist in a vacuum. It is part of a broader, messy scramble in the payments space. Everyone is trying to figure out how to verify agent intent and capability. We are seeing tools like Mastercard’s probability scoring attempt to solve this from the identity side. Meanwhile, the actual integration of these agents into the shopping experience – via the Personal Agent Protocol or platforms like Shopify Muse – requires a rock-solid settlement and dispute resolution mechanism. AmEx is trying to provide that, but they are playing catch-up with the reality of how these agents actually behave in the wild.
There are still plenty of open questions. How exactly does AmEx define a ‘registered’ AI agent? What are the technical hurdles for registration? Will this protection cover third-party agents, or is this just a walled-garden play for the AmEx ecosystem? And since this is currently a US-only announcement, the global merchant community is left waiting for a real standard. AmEx has correctly identified the primary friction point in agentic commerce, but moving from an advisory playbook to a functional, risk-managed reality is going to take a lot more than just good intentions.