Agentic Commerce Wars: Win Trust, Take It All

The hottest question around AI agents isn’t whether they’ll soon shop, book, or sign contracts for people. They’re rapidly getting good at that. The real question is who can credibly prove that an agent acted for a real user, under a legitimate instruction, and without being manipulated. That’s exactly where the new push from the FIDO Alliance comes in. And with it, the focus in agentic AI is shifting from capability to accountability.

At the end of April, the FIDO Alliance announced plans to develop new standards for trustworthy agentic interactions and commerce. It’s launching an Agentic Authentication Technical Working Group and drafting specs for agent-initiated commerce. The effort builds partly on contributions from Google and Mastercard, including Google’s Agent Payments Protocol and Mastercard’s Verifiable Intent framework. The goal is simple to state: AI agents must be able to act on users’ behalf safely, privately, and interoperably.

The backdrop is fundamental. The internet was built on the assumption that a human presses the button. AI agents break that assumption. Once an agent logs in, initiates a payment, or changes an account on behalf of a user, classic authentication isn’t enough. You also need to capture intent, delegated authority, and traceability. FIDO says it plainly: without new standards for intent, agent identity, and authorization, adoption of this entire category stalls.

Because the market knows agentic commerce doesn’t scale without a trust layer. A consumer will only let an AI agent buy something if it’s clear what the agent may and may not do, how that’s provable, and who’s liable when things go wrong. The same goes for merchants and banks. They don’t want an ecosystem where everyone argues after the fact whether a purchase was “really” intended by the user. That makes open standards suddenly more strategic than yet another model upgrade.

We’ve seen this pattern in past internet waves. First comes the demo. Then commercial interest. Only then do protocols, certificates, and control layers emerge to run at scale. Agentic AI is entering exactly that phase. Earlier AIwereld reporting on machine payments via Stripe and on agentic AI as a market layer makes this step logical: the infrastructure for the agent economy is starting to take shape.

Open standards are often underrated in AI coverage. They’re not sexy, but they decide who can dominate the market. Whoever writes the trust and authorization layer ultimately determines how easily agents move between platforms, merchants, and payment rails. That’s a more powerful position than just shipping a smart model. In that sense, this moment looks more like the history of payment networks and identity standards than that of chatbots.

It also hits security head-on. AIwereld has covered prompt injection in agents. An agent vulnerable to manipulation that can still make payments or change accounts is a brand-new risk profile. FIDO’s initiative aims to close that gap—not by making agents less autonomous, but by locking down their permissions cryptographically and procedurally. Trust stops being a UX layer and becomes a security architecture.

For companies, the message is clear: don’t start agentic commerce by picking a model—start by choosing a trust layer. Any retailer, bank, platform, or software vendor that will admit agents must define how authority is delegated, what limits apply, how transactions are logged, and who handles dispute resolution. That’s not a detail for later. That is the product design.

Europe has an opening here. If the EU is serious about privacy-preserving identity and age verification, it can also shape standards for agent authorization. If it doesn’t, the rules of agentic commerce will again be written by American platforms, payment networks, and software firms. The question of who trusts the agent is ultimately the question of who writes the digital market standard.

The agent economy won’t be built on smarter agents, but on stronger guarantees. FIDO’s move highlights the real bottleneck: not intelligence, but trust. The next AI battle won’t just be about models—it’ll be about protocols, identity, and control.