Stripe is introducing scoped credentials, agent-specific API keys and approval rules that let AI agents handle payments while keeping sensitive transfers under human oversight.
Stripe has introduced a set of controls designed to let AI agents move money while keeping humans in charge. In a blog post, the company said its Model Context Protocol server can already help agents build checkout pages, run financial reports and automate refunds or price changes, and can now also send funds through Stripe Treasury. The new approach is built around two operating modes: an agent can act on a user’s behalf, or it can be given its own identity with tighter oversight.
When an agent connects through OAuth and acts on behalf of a user, Stripe says the system can be scoped to specific accounts or sandboxes and limited to read or write access depending on the task. For sensitive live-mode actions such as refunds or payouts, Stripe requires a human to confirm the request before money leaves the account. The company says that safeguard remains in place even if the client has been configured not to ask for permission on every tool call.
For teams that want automation to run without tying every decision to a single person’s login, Stripe is also offering agent-tagged API keys. According to the company, these keys give an agent long-lived access that is independent of any individual team member, while still supporting permissions, rotation and expiry. Stripe says the keys are also subject to approval rules, so designated reviewers can be required to sign off on sensitive actions such as refunds, payouts and outbound transfers. The company says these default rules apply to agent keys from the outset, and can then be tuned to set thresholds, choose reviewers and decide whether an action should be approved or blocked.
Stripe says setting up an approval rule takes only a few minutes in the Dashboard. The process involves choosing a trigger, such as a refund being created, setting a condition such as amount, selecting whether the rule applies only to agent-tagged keys or to all actors, and naming the reviewers who can sign off. If an agent hits a rule, the API returns an approval_required error and opens a request for review. Once a designated reviewer approves, Stripe attempts the action again; if the request is denied, nothing goes through. Requests that remain unresolved for 14 days expire automatically.
The company also says agents can be kept informed through webhooks that report whether a request was created, approved, rejected, cancelled, expired, succeeded or failed. Taken together, Stripe argues that scoped credentials, human confirmation and configurable approvals let businesses decide which actions can happen automatically and which require a second set of eyes. The pitch is that teams do not have to choose between full automation and manual control: they can start with confirmation for agent actions, then add two-party approval once the agent becomes a more central part of operations.
Disclaimer: This article is intended to inform and educate, not to recommend or endorse any financial product, investment or strategy. Please consider your own financial circumstances and seek professional advice where appropriate before making financial decisions.
