Permissions
Checks whether the agent has more access than its task requires.
Plain-English AI security check
Email Agent Risk Assessment
Check inbox access, prompt injection, impersonation, credential scope and autonomous sending risk.
Example aggregate declared score46/100
Moderate overall declared band
What the assessment covers
Untrusted input
Reviews how external content can influence tools and actions.
Autonomy
Measures approval gates, limits and potential blast radius.
Evidence
Evaluates logging, testing and incident containment.