Stripe's Link Wallet Lets AI Agents Spend on Your Behalf - Without Exposing Your Cards

Stripe has launched a significant expansion of its Link wallet service, allowing AI agents to make purchases on behalf of users whilst maintaining strict security controls. The payment giant's 250 million-strong user base can now grant OAuth access to AI agents, enabling them to request one-time-use cards or Shared Payment Tokens backed by existing stored payment methods. The system's architecture prioritises security and user control. AI agents never gain access to underlying payment credentials, and every transaction requires explicit user approval through real-time notifications. This approach positions Stripe as the consent layer between AI agents and merchants, rather than simply offering another checkout solution. Initial compatibility includes Claude, OpenAI's models, and custom-built agents. Stripe has indicated that support for emerging payment methods, including agentic tokens and stablecoins, will follow. The service is rolling out first in the United States before expanding to other markets. The practical implications remain to be seen. Whilst the security framework addresses concerns about AI agents handling financial credentials, the requirement for manual approval of every purchase may limit adoption for high-frequency transactions. Users may find themselves inundated with notifications for routine purchases, potentially undermining the convenience that AI agents are meant to provide.