Executive Summary
Token reports building a browser agent that broke its own security rules by default, including independent identity, job-appropriate access, and separate traceability. Organizations deploying AI agents should review identity, authorization, and audit controls.
What Happened
We tell customers three things about securing AI agents: they need their own identities, access that fits 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. This helped us think through how such agents should be defended. When a browser agent acts inside a user’s authenticated
Affected Products
- Token browser agent
Vulnerability / Attack Technique
Excessive agency through authenticated user context
Technical Mechanism
The article states that the browser agent acts inside a user’s authenticated context. This caused the agent to lack its own identity, receive access not explicitly fitted to its assigned job, and rely on traceability associated with a human rather than being separately attributable.
AI Security Category
LLM06:2025 Excessive Agency
Part of the AI Agent Excessive Agency topic.
Relevant Frameworks
RealGround Analysis
Agents operating within authenticated user sessions can blur the distinction between human and machine actions and may inherit broader permissions than their task requires. RealGround recommends validating agent business logic, enforcing isolated identities and least privilege, and red-teaming authenticated browser workflows.
Recommended Controls
- Assign each AI agent a distinct non-human identity rather than using a human user’s identity
- Enforce task-specific least-privilege permissions for every agent action
- Maintain immutable audit trails that separately attribute actions to the agent and the human who initiated the workflow
- Test authenticated browser-agent workflows for unauthorized actions and permission overreach
- Require explicit approval or policy checks for high-impact actions performed in user sessions
Related RealGround Service
AI Agent Business Logic Audit · or start with the free AI Agent Security Scanner.
Sources
thehackernews.com: https://thehackernews.com/2026/11/we-built-agent-that-broke-all-of-our.html
Summary and analysis written by RealGround from the source above; last updated 2026-11-12.
