Privacy.
Built in, not bolted on.
Your data is scoped to your organisation. AI only sees what a task needs. And the decisions that matter always wait for a person.
AI governance
Applies to all 18 appsProcessing
Data sent to models
Always on
Illustrative admin screen. Self-hosted AI is a planned option.
Four commitments
Tenant isolation.
Every table is tenant-aware. An organisation’s users can only ever reach that organisation’s data — and AI inherits the same limits.
A controlled AI gateway.
No app talks to a model directly. Central rules decide what may be sent, strip what isn’t needed, and record what happened.
People approve.
Financial, legal and hiring actions always need an authorised reviewer. AI prepares; it never decides.
Everything is recorded.
Requests, approvals and processing are kept as immutable audit events under your retention policy.
The AI gateway
Every model call takes
the same road.
- Step 1: RequestAn app asks for help.
- Step 2: Policy checkAllowed for this tenant and role?
- Step 3: MinimiseOnly the fields the task needs.
- Step 4: Approved modelYour chosen processing option.
- Step 5: ValidateSchema and rule checks on output.
- Step 6: Human reviewHigh-impact actions wait.
- Step 7: AuditRequest, decision, result.
An honest distinction: “not used for model training” does not mean “never sent to an AI provider”. With managed AI, selected data may be processed externally under contracted terms. We’ll always tell you which.
Human approvals
Who signs off on what.
Controls we test in every pilot
- Tenant isolation tests
- Access revocation
- Encryption in transit and at rest
- Retention and deletion
- Incident handling
- Data export on request
- Malware scanning of uploads
- Immutable audit events
Designed to align with ISO 27001:2022 and SOC 2 controls.
Control design follows these frameworks. Certification scope and timing for SmartAtom are still to be confirmed — we won’t claim them before they exist.