Why is this big news? Once one network says “we cover agent errors,” the others can’t say no.
The Problem We’ve Been Waiting for a Network to Solve
For the past eighteen months, I’ve written extensively about agentic commerce as a test of *incentive alignment*, not technology. The tech works. What doesn’t work is getting all parties—networks, issuers, merchants, platforms, and payment processors—to align around who owns the agent, who owns the data, and who bears the risk.
Today, American Express did something important: it solved that problem for its own closed loop (and its customer base). What does this mean? I hope it means US Issuers will lean in on the V/MA solutions that can allow them to operate at near parity (V/MA have the rules, tech and governance). But changing a network is really hard.
What Amex Announced
American Express unveiled two interconnected products:
1. **Agentic Commerce Experiences (ACE) Developer Kit** — Tools that let technology companies, merchants, and platforms build agent-powered transactions with agent registration, account enablement, intent analysis, and payment credential verification.
2. **Amex Agent Purchase Protection™** — The network is now extending its legendary purchase protection to cover transactions initiated by registered AI agents. If an agent makes a mistake, Amex covers it. “We’re standing behind the agents,” said Luke Gebb, EVP of Innovation at American Express.
This is industry-first coverage. Visa has the framework and the rules (DAF, TAF, Intelligent Commerce, DCAP, …et) . Google has the protocols and the platform. But neither has yet said: *we will put our balance sheet behind purchases made by agents.*
Why This Breaks the Deadlock
Here’s the incentive structure that’s been stuck:
- **Merchants** want agents to drive conversion and eliminate checkout friction, but they’re scared of getting locked into new VAS with TBD pricing, data sharing and agent-driven fraud.
- **Consumers** want to delegate shopping to agents, but they’re scared agents will make mistakes or buy the wrong thing.
- **Networks** want agent commerce to scale, but they’re constrained in getting banks to take ownership of liaiblity for a solution they don’t control. Similarly, if banks COULD deliver a solution w/ 100% auth rate and 0% fraud (i.e., perfect auth), it would be a nightmare to everyone who has invested in building a device graph.
- **Processors** want to retain their competitive advantage and extend into the plumbing of agentic and onboard merchants in front-end MCP/UCP flows, but they can’t move faster than the networks allow.
Amex 3 Party Advantage
By extending purchase protection to agent transactions, Amex is saying: “We will absorb the asymmetry of information and the tail risk so that agents can actually work.”
Amex, as a 3-party network, doesn’t have that friction. Because they are both the issuer and the network, they have total control over the credential, the message, and the participant. This allows them to do something truly “industry-first”: Amex Agent Purchase Protection.
From the Amex PR:
“In the future, if a Card Member authorizes an AI agent to make a purchase and that agent sends American Express the customer’s authenticated purchase intent, American Express will protect eligible customers from charges related to AI agent error.”
This directly addresses the “hallucination” risk I discussed in my analysis of agentic liability, where the decoupling of intent from action creates a “Trust Vacuum”.
The Google Win and the Standard War
In my view Google’s support is also big news. Getting Google to align with a network-led authentication model is a massive strategic victory. As I’ve noted, Google’s AP2 and “Buy for Me” initiatives represent the most significant growth opportunity for networks, but only if we can bridge the gap between Google’s x509-based intent and the network’s settlement rails.
By securing a commitment that integrates with Google’s agentic ecosystem, Amex is proving that identity needs a network to scale across domains.
A Contrast in Strategy: PAZE and PayPal
It is hard not to feel the frustration of V/MA here. Both have the tech, rules and governance (like Visa’s Intelligent Commerce, TAP, DAF, DCAP, …etc) but they are fighting a divisive war for control with their own issuers. For example MA pushed agentic mandates into their network, but Google has no clue that it requires every Issuer to buy in and accept them. MA has no power to have an Issuer accept a Google cryptographic chain. Their daily MA success stories are lab demonstrations with Bank Innovation teams.
Look at PAZE, which officially “went live” today after over two years in pilot. Despite the effort, PAZE still struggles to offer a clear merchant benefit over established giants like ApplePay or ShopPay. You can check their current merchant directory here, but it highlights the difficulty of building a new brand from a bank consortium versus the “cleaner” execution of a 3-party model.
I also find myself wondering about PayPal. They have a 3-party model, but their current “peace agreements” with V/MA largely constrain them to operate within the 4-party rules, limiting their ability to innovate as freely as Amex just did.
## Connecting to Where We’ve Been
If you’ve read Agentic commerce economics, you know the core insight: agentic transactions represent a purchase order with a payment instrument if done correctly. This collapsing of the funnel has been put on hold by OpenAi and other major retailers because of the problem Amex (and Google) is solving here.
The authentication piece matters too. As I’ve discussed in Google’s advantage in passkeys and credentials, the network that can bind device-level trust to agent-level authorization wins. Amex’s ACE kit is building exactly that—intent analysis, credential verification, and app-level controls that let consumers say “yes, Agent X can buy if the price is below $300.”
And this connects to the deeper point I’ve made about [AP2 and the limits of protocol innovation](https://blog.starpointllp.com/2025/10/stripe-agentic-commerce-protocol-acp/): protocols are necessary but not sufficient. Someone has to stand behind the transaction. Today, that’s Amex.
## The Next Move
Amex just raised the bar. Issuers need to LEAN IN on the network models created by Visa, Mastercard to match this—or explain why they won’t. Consumers will demand protection. Merchants will demand it. Once one network says “we cover agent errors,” the others can’t say no.
The real test: will the other networks’ closed loops (Visa’s issuer business, Mastercard’s new models, Block’s square/cash ecosystem) enable them to offer the same protection? And if they can’t, what does that mean for their place in agentic commerce?