Security
Security belongs where intent becomes execution.
Model safeguards are not the same as authorization. Aegisora places deterministic enforcement between agent intent and execution.
Runtime decisionsLive
- ALLOW
crm.read_ticket
support-agent · Within scope
- ESCALATE
files.export salary.csv
hr-agent · Sensitive data class
- BLOCK
customers.delete bulk
billing-agent · No authorization
Controls
Four ideas behind the boundary.
Canonical identity
Know which agent requested an action.
Restricted tools
Block capabilities outside the allowed execution surface.
Pre-execution gate
Make the control decision before execution is attempted.
Evidence trail
Keep the enforcement result reconstructable.
Review how the boundary is protected.
Read the trust documents or talk to the team about a security review.