security_and_control
Agents should be useful before they are autonomous.
We design agents with clear boundaries, approval points, access controls, and escalation paths so they can remove repetitive work without creating unmanaged risk.
level_01
Observe
The agent reads and summarises only.
level_02
Advise
The agent recommends actions; humans execute them.
level_03
Act with approval
The agent prepares actions and asks for explicit human approval.
level_04
Act within guardrails
The agent performs limited actions inside agreed rules, with logs and escalation paths.
control_principles
How Nusa Intelligence keeps humans in control.
Human approval
Sensitive actions always require explicit human approval.
Least-privilege access
Agents get only the access they need — no more.
GDPR-aware setup
Designed around lawful basis, minimisation, and clear data flows.
Audit logs
Logs where possible so every action is reviewable.
Escalation paths
Uncertain or risky cases route to a human owner.
Role-based access
Different roles see different things — by design.
No unnecessary exposure
Private knowledge stays private.
Vendor selection
Tools chosen for risk profile, not hype.
What agents won't do
No payments, contracts, or sensitive comms without approval.