AI agents that can browse the web, fill out forms, and place orders on your behalf are coming fast. The security infrastructure to keep those agents from making unauthorized purchases? That's lagging behind.
The FIDO Alliance - the industry body behind passkey authentication standards - is partnering with Google and Mastercard to develop standards for how AI agents authenticate and authorize financial transactions. According to Wired, the goal is to create a framework that lets an AI act on your behalf without handing it unconstrained access to your payment methods.
The core problem is one of delegation: when you tell an AI agent to "book the cheapest flight to Chicago under $400," how does your bank know the agent is acting within those constraints and not going rogue the moment a better deal requires a different credit card? Current payment systems weren't built to handle machine-to-machine authorization at that level of specificity. Credit cards have spending limits, but not intent limits.
FIDO's existing work on passkeys - the password-replacement technology now supported natively by Apple, Google, and Microsoft - gives them real credibility to shape these standards. Mastercard brings the payment network perspective. The open question is whether whatever they produce will be widely adopted before AI agents become mainstream shopping tools, which on current timelines means within 12 to 18 months.
Who Carries the Liability Right Now
For businesses selling online, this ambiguity is already a practical concern. Every major AI assistant is building agentic features: ChatGPT has shopping integrations, and Claude's computer use capabilities can operate browsers and complete checkout flows. Without standardized authorization protocols, merchants and consumers face murky liability when an AI agent makes a purchase that wasn't exactly what the user intended - or worse, when a compromised agent drains an account.
The FIDO/Google/Mastercard initiative is the right conversation to be having, even if the resulting standards are two or three years from being embedded in actual checkout systems. The window between when AI agents go mainstream and when secure payment rails exist for them is the danger zone.