Least privilege
We begin with the minimum access the workflow needs and use read-only connections where possible. Any ability to change a source system must be explicit.
Trust and delivery
The controls are part of the system.
High-trust AI is defined by the boundaries around real work: who can act, who must approve, what remains inspectable and how failure is recovered.
Human approval
Operators can inspect, approve, correct or stop the system at the points where consequence demands judgment.
Inspectable work
Important inputs, outputs, approvals and exceptions remain visible enough to understand what happened and why.
Recovery
Failure modes, escalation paths and operational ownership are defined before a deployment enters real work.
Data handling
Data is encrypted in transit and at rest. We define data handling and responsibilities for each deployment to meet applicable Singapore PDPA requirements. Data sources, providers, hosting, retention and deletion are documented for the deployment.
Ownership and handover
Kairos supports the system after launch. Client data, code, documentation, support and ongoing operation are agreed in scope so there is no ambiguity at handover.
Before build
Make the data path explicit.
Document the answers for your workflow, its data and the decisions it supports.
- 01
What information enters the workflow?
- 02
Where is it processed and stored?
- 03
Which people and systems receive access?
- 04
Which actions require human approval?
- 05
What is recorded, retained and recoverable?
- 06
Who owns the system after deployment?