What security looks like when you engage IntraWork
When a prospect asks us "what does security look like?", they are usually asking three different questions at once: what does the engagement actually involve from a setup standpoint, where does the data physically sit, and what stops something going wrong. This overview answers all three in two pages. Detail is gated behind the rest of the security pack and, where it is genuinely commercial-in-confidence, behind an NDA.
1. What we set up for you
The IntraWork Platform is a per-customer self-hosted deployment. That phrase reads, on first encounter, like it puts the burden on your IT team to spin up infrastructure. It does not. Per-customer self-hosted means your data lives in your own infrastructure under your own contractual relationships with the underlying vendors — and we do the setup work on your behalf as part of onboarding.
Specifically:
- We stand up your Supabase deployment in the region you select (default
ap-southeast-2Sydney for Australian customers) and configure the database schema, RLS policies, and authentication. - We register the Anthropic API account in your name where you do not already have one, and transfer ownership and billing to you as part of go-live.
- We configure the Microsoft 365 integration. Your operator authorises via OAuth; tokens land in your deployment.
- We deploy the Next.js application to your dedicated Azure Static Web App.
Your IT department is welcome to take this on themselves where preferred. The default, however, is that we handle it — onboarding is a service, not a self-serve. Self-hosted means yours. Service-delivered means easy.
2. Where your data lives
Three places matter:
- Your operational data — contacts, projects, agreements, agent activity, brain content — lives in your own Supabase deployment, in the region you select. We do not aggregate this data across customers. There is no shared multi-tenant database. Tenant isolation is physical, not logical.
- Your Microsoft 365 data — mail, calendar, OneDrive, Teams — lives in your existing Microsoft 365 tenant. The platform reads it via Microsoft Graph on-demand, under operator-delegated OAuth, with minimum-necessary scopes. We do not copy your mailbox or document store wholesale to our infrastructure.
- Your agent prompts and responses travel to and from the Anthropic API in the United States. Anthropic's Commercial Terms prohibit training on customer data submitted via the API. We pin model versions per agent.
IntraWork as a company holds no customer operational data. We cannot, by architecture. The complete sub-processor list — Supabase, Microsoft, Anthropic, Azure Static Web Apps, and Wix where applicable — is published at /legal/sub-processors.md.
3. Who can access what
Three access classes, ranked from broadest to narrowest:
- You and your authorised staff. Authentication via Microsoft OAuth with multi-factor authentication enforced at the Microsoft 365 account layer. Additional Supabase TOTP enforcement applies on administrative routes. Role separation between operator and client-portal users is enforced inside the application via Supabase RLS.
- IntraWork (Jessica). No standing access to your deployment. Support access is granted by you, time-limited to the engagement, scope-limited to the work agreed, and audit-logged.
- Sub-processors. Each sub-processor operates inside its own service boundary under its published terms. Supabase sees your operational data at rest; Microsoft sees your M365 data; Anthropic sees prompts and responses for the lifecycle of each API call.
The architectural primitive that backs all of this is the approval queue. Every agent action that touches the outside world routes through an approval queue inside your deployment before any outbound send. The operator reviews, edits, approves, or rejects. No agent action reaches a third party without an explicit human approval.
4. What happens if you leave
The exit is built in.
- A 30-day grace period applies from the date you give notice. During this window the deployment continues to operate and you continue to have full access.
- We provide a complete data export in standard formats — JSON for relational records, CSV for tabular data, Markdown for brain content. The export is yours to keep.
- After the grace period, your operational data is permanently deleted from any infrastructure IntraWork operates on your behalf.
- Your own Supabase deployment, your own Microsoft 365 tenant, and your own Anthropic API account remain yours. You decide whether to keep them running, migrate them to a different platform, or wind them down.
- IntraWork holds no copy to delete because IntraWork never held a copy. The deletion obligation is on your underlying vendor accounts, not on us.
The full mechanics are in the Data Processing Agreement template, which becomes part of your engagement agreement.
5. Where we are honest
We do not currently hold SOC 2, ISO 27001, ISO 42001, IRAP, or HIPAA certifications. The honest interim position:
The IntraWork Platform is architecturally aligned to industry control frameworks — SOC 2 Type II controls, OWASP Top 10 for LLM Applications (2025), and the ISO/IEC 42001 AI management system control set. Our Information Security Policy and Incident Response Runbook describe how those controls are implemented.
Third-party certification (SOC 2, ISO 27001, ISO 42001, IRAP) is deferred until customer demand and engagement scale justify the cost — typically AUD $30,000–$60,000 per audit cycle for a business of our size. We will pursue certification when a specific customer engagement requires it as a contractual condition, with the customer participating in audit cost where appropriate.
A 38-item internal security self-audit aligned to CSA CAIQ, SIG Lite, and OWASP LLM controls is maintained. Our current score is 32/38, with the remaining six items on a scheduled remediation roadmap.
An independent penetration test will be scheduled within 12 months of first paying customer engagement. The remediation plan will be tracked transparently.
For procurement teams who need a full CAIQ v4 or SIG Lite questionnaire response, that documentation is available under NDA.
The architectural choice — per-customer self-hosted with physical tenant isolation, operator-in-the-loop by design, no model training on customer data — is the security posture we lead with. It is a stronger posture than the certification stamps alone would imply, and the rest of the security pack documents it in detail.
For further detail, see the trust page, the privacy policy, the sub-processor list, and the DPA template. For a security questionnaire response, contact security@intrawork.com.au.