browser-agent-security

What problem does it solve?

Browser-controlling AI agents expose a remote-code-execution surface when interacting with web content, making them vulnerable to DOM prompt injections, credential harvesting, and session hijacking.

Core Features & Use Cases

• DOM sanitization and untrusted content labeling to prevent prompt injection by web content entering the agent's reasoning context.
• Credential isolation with a dedicated manager that fills forms without exposing credentials to the agent, enforcing domain checks and user confirmation for submissions.
• Session and domain isolation to ensure credentials and session data are scoped to the current task and domain, reducing leakage risk.
• Operational guidance for reviews across typical browser automation tasks, including alerting on injection patterns and enforcing human-in-the-loop for critical actions.

Quick Start

Run a browser agent security review on a given web task to identify DOM injection and credential leakage risks.