Token Explains Identity and Access Risks in Browser Agents

Summary
Token says its Rehearsal browser agent exposed identity, scope and audit gaps, and recommends discovering agents using human sessions, constraining access and requiring step-up approval for sensitive actions.
Key points
- Rehearsal performs tasks in a user's browser session, so applications may see the user's identity without knowing which agent is acting or what its task permits.
- Token says its control of both Rehearsal and its platform allowed it to identify the agent and apply scope and page-level controls unavailable to outside agents.
- The article describes browser controls, WebMCP tools and Web Bot Auth as partial approaches; it says they do not yet jointly convey agent identity, human delegation and task scope.
- Security teams are advised to discover agentic browsers and extensions, identify whose sessions they use, and add human approval for sensitive actions, such as with a passkey prompt.
- Until identity, delegation and scope can be conveyed together, the article recommends treating agents operating in human sessions as delegated access that must be discovered, constrained and audited.
Article Details
- Topic
- Security controls for AI agents operating in users’ authenticated browser sessions, including agent identity, delegated scope, authorization, and auditing.
Vendors
GoogleThe app can publish its actions as tools. WebMCP, a draft that Google and Microsoft are writing in a W3C community group, lets a site expose actions that an agent can call directly and flag each one as read-only orMicrosoftits agent to the sites the task requires, and tags the pages the agent visits in the user’s history. Microsoft lets admins turn agentic browsing on per site and pause it before passwords and card numbers are entered.Tokenfits the job they’re given, and traceability that’s separate from any human’s. That was true for us at Token as well. Then, we built an agent that broke all three by default.
Products
Edgedon’t, and none of them passes it on. The browser knows its agent is driving, but neither Chrome nor Edge describes how to tell the app on the other end. A WebMCP tool runs inside the user’s session, with theGoogle ChromeThe browser can mark when its agent is driving and restrict itself accordingly. Chrome auto-browse, for example, already requires a person’s approval before high-stakes actions such as payments or mass emails, limitsRehearsalThe agent we created is named Rehearsal. Its job is to answer questions inside the Token platform by doing rather than telling. It autonomously opens pages in the browser tab, sets filters, and fills in forms.Token platformThe agent we created is named Rehearsal. Its job is to answer questions inside the Token platform by doing rather than telling. It autonomously opens pages in the browser tab, sets filters, and fills in forms.
