Decision Lab
See BehalfID stop unsafe AI actions before they run.
Choose an AI action, run it through a permission passport, and see whether BehalfID allows it, blocks it, or pauses for human approval.
Research Agent
Reads a public documentation page
Agent request
- agent
- Research Agent
- action
- browse_web
- vendor
- web
- resource
- web
- route
- /docs/getting-started
Matching ruleallow browse_web on web
The active permission explicitly allows public web browsing. No checkout, form submit, or credentialed request is in scope.
DecisionAllowed
Allowed within scope
The read-only tool can run.
Execution
Tool ran.
Audit log
Allowed decision logged before execution. verification.allowed
What this proves
Prevents unsafe actions
Tools do not run unless the agent has permission.
Preserves useful automation
Safe actions can proceed without constant user interruption.
Creates an audit trail
Every allowed, blocked, or approval-required action is recorded.
Add the decision point.
This sandbox is simulated. No real backend permissions, network requests, or agent actions run here.