Least-privilege access
Automations should receive only the systems, records, and actions required for their defined role.
Security & governance
Useful AI needs clear boundaries.
Kenzai designs security, access, review, and observability into the workflow itself. The goal is not autonomy at any cost; it is dependable automation with accountable control.
01Context
02Policy
03Human review
04Approved action
05Audit signal
Control model
Security is an operating design problem.
Controls are strongest when they match the real workflow: who can see what, which action is allowed, where review is required, how errors surface, and how access is removed.
Human review points
Material commitments, sensitive changes, and ambiguous exceptions remain visible to an accountable reviewer.
Traceable activity
Important inputs, recommendations, approvals, actions, and exceptions should be observable enough to investigate and improve.
Data discipline
We minimize unnecessary data movement, separate environments where appropriate, and avoid placing secrets in client-side code.
Deployment review
Controls are defined before production access.
Data sources and sensitivity
Credentials and permission scope
Allowed and prohibited actions
Human approval and escalation points
Logging, monitoring, and failure handling
Retention, removal, and offboarding
