At Stripe Sessions in April 2026, the company introduced the Link Agent Wallet, a feature designed to bridge the gap between autonomous AI agents and the actual movement of money. At the heart of this infrastructure are Shared Payment Tokens (SPTs), which represent the first attempt at a consumer-side trust primitive for agent commerce. While the industry has spent months debating the plumbing of agentic transactions, Stripe is now attempting to address the human side of the equation: the fact that most people are still deeply uncomfortable letting software spend their money.
Technically, SPTs function as scoped clones of a customer’s existing payment method. Rather than handing an agent a blank check, a user can configure specific constraints: currency limits, maximum transaction amounts, and expiration windows. The agent operates within these guardrails, and the consumer retains the ability to revoke access at any time. Sellers, meanwhile, receive webhook events detailing consumption, expiration, and revocation. The system supports a range of payment types, including cards, Klarna, Affirm, and stablecoins via MPP. As of July 2026, however, the technology remains in preview under API version 2026-04-22.preview.
Stripe is leveraging its 250 million Link users to determine if these granular controls can overcome deep-seated consumer hesitation. The problem they are trying to solve is starkly illustrated by an April 2026 survey from Product.ai, which found that only 14% of consumers trust AI to execute purchases without manual verification. Furthermore, 42% of respondents stated they would not trust an AI for purchases above a certain threshold, though the survey did not specify the exact monetary value of that limit. This 14% trust ceiling is the primary barrier to the widespread adoption of agentic commerce, and SPTs are the industry’s first formal attempt to build a technical solution to that psychological hurdle.
During the Sessions event, Stripe leadership emphasized the necessity of this trust layer. John Collison noted the complexity of the challenge, stating:
Agents need to be able to pay, businesses need to be able to accept payments from those agents, and the whole system needs a trust layer, and Stripe is working hard and getting all of this deployed.
This sentiment was echoed during a trust panel at the same event, where speakers highlighted the fundamental anxiety consumers feel regarding autonomous systems:
consumers will say, Can I trust this? Are things going to go well? And if they don’t go well, what’s going to happen? That’s an element of bringing trust to consumers.
To mitigate these risks, Stripe is applying its existing fraud detection infrastructure, Radar, to agentic transactions. The company claims that SPT-based payments undergo the same level of scrutiny as direct checkout flows. It is a logical move, but it highlights the persistent tension in this space: the infrastructure is rapidly maturing, yet consumer behavior remains cautious. We have seen this pattern before, notably in the x402 volume mirage where 50% of reported activity was artificial, leaving only $28,000 in real daily volume, and in the Three-Rail War, where competing rails focused on Category 1 and 2 plumbing rather than consumer-facing commerce.
Despite the technical sophistication of the Link Agent Wallet and the promise of SPTs, there is a notable lack of data regarding their real-world performance. As of July 9, 2026, no adoption metrics have been published. We do not know how many users have actually configured an agent, nor do we know if the existence of these controls is enough to push that 14% trust figure higher. The regulatory environment also remains ambiguous; while the GENIUS Act addresses stablecoin issuers, there is currently no framework for AI agent-initiated transactions, and the liability model for when an agent makes an unauthorized or erroneous purchase remains unresolved.
For now, SPTs exist as a preview-stage tool—a piece of infrastructure waiting for a market that is still largely skeptical. The mechanics of scoped, revocable tokens are sound, but they are only one part of a much larger, unproven ecosystem. Whether these primitives can successfully lower the barrier to entry for agent commerce, or if they will remain a niche feature for early adopters, is a question that the current lack of adoption data leaves unanswered. The industry has provided the technical framework for agent-initiated payments, but it remains unclear if consumers will adopt these tools for routine financial activity.