Security & Privacy
Aegis Eight only reads the security and configuration information required to assess Essentials compliance from your connected cloud tenant — Microsoft 365 or Google Workspace. It cannot modify your tenant and cannot access email, files, chat, or other user content. The rest of this page explains exactly how, in plain English — no marketing copy.
Last reviewed 2026-07-14. The canonical source-controlled version of this document lives in docs/security.md in the Aegis Eight repository.
Choose your cloud platform for connection method, access boundaries, revocation steps, and evidence examples. Data handling, isolation, and compliance sections below apply to both.
What Aegis Eight can access
- Security and policy configuration (e.g. Conditional Access, MFA policies)
- Which MFA and authentication methods are registered
- Device compliance and operating-system version state
- Admin roles and privileged-access configuration
- Microsoft 365 Backup protection and restore evidence
What Aegis Eight cannot access
- Your email or mailbox content
- Your files in OneDrive or SharePoint
- Your Teams messages
- Passwords or MFA settings — we cannot change anything
- Any tenant other than the one that granted consent
How a scan works — Microsoft 365
No configuration changes are ever made to your tenant.
1. Trust model — what consent actually grants
Microsoft Graph is Microsoft’s official API for securely accessing Microsoft 365 configuration and security settings.
Aegis Eight is a multi-tenant Microsoft Entra application with Application permissions (background, daemon-style access). When a Global Administrator grants admin consent, a service principal of Aegis Eight is provisioned inside their tenant. From that point on the worker authenticates with the OAuth2 client_credentials grant — server-to-server, no user context, no session, no impersonation.
| Aegis Eight has | Aegis Eight does not have |
|---|---|
| Read access to the specific Graph resources listed on /permissions | Write, create, update, or delete access to any Graph resource |
| Access to the tenant whose Global Administrator granted consent | Access to any other tenant — multi-tenancy means we are deployed to many tenants, not that we read across them |
| The ability to enumerate users, devices, and roles for compliance scoring | Mailbox content, OneDrive files, SharePoint files, Teams chats, calendar entries, or any user content |
| The ability to act in the tenant for as long as consent persists | The ability to act after consent is revoked — Microsoft enforces this at the token-issuance layer |
2. Exactly what we access, and why
The authoritative list is on /permissions and matches the Microsoft consent screen exactly. The pattern Aegis Eight commits to:
- Least privilege.A scope is added only when a collector that needs it ships. Scopes are never reserved “for future use”.
- Read-only. No *.Write.* or *.ReadWrite.* scope appears anywhere in the request set.
- Mapped to a control. Every scope appears next to the Essentials strategy it supports.
- Honest parity on Google. Optional add-on modules use separate Entra app registrations — never bundled with the core scan without their own consent step.
Collectors source MFA registration, Intune compliance, directory roles, Conditional Access policies, and Microsoft 365 Backup summaries from Graph.
Every permission Aegis Eight requests is read-only, and there is no write surface anywhere in the product. Any future feature requiring write access would be introduced through a separate, explicit consent process.
5. Evidence chain — provenance, not just findings
The reason this section exists is the January 2026 IRAP Quality Assurance Framework and the parallel insurer expectation of enforcementevidence. A report that says “MFA registration is at 98%” without showing the underlying Graph response, the timestamp, and a hash of the bytes is increasingly hard to defend.
We preserve a verifiable evidence chain: every Graph response that informs a finding is captured, SHA-256 hashed, and linked to the finding. An assessor can re-query Microsoft Graph (within retention windows) and confirm the data matches exactly what we scored — without taking our word for it.
- Provenance on API-derived findings. Each carries an
evidenceSourcewithqueriedAt(UTC),endpoint(the exact URL queried),responseHash(SHA-256 of the response body), andsnapshotRef. - responseHash is the integrity claim. Even if storage is re-keyed or migrated,
responseHashlets the customer confirm whether a stored body matches what Aegis Eight scored. - Tier discipline is structural. ASD evidence tiers are derived from a meaning-shaped
evidenceTypediscriminator in code — not picked per-finding by a human. - The roadmap is honest about today’s tier. Current Graph-snapshot collectors land at the Fair tier; Path 2 collectors lift specific findings to Good. Synthetic-test (Excellent) is sequenced post-MVP for Microsoft 365.
Every per-finding evidence cell in the PDF cites the live endpoint queried, the UTC timestamp, the responseHash prefix, and a snapshot reference.
7. Customer control
A Global Administrator can fully sever the relationship at any time:
Revocation takes effect immediately at the Microsoft side; the next scan attempt will fail with an authorisation error and the dashboard will surface a “we lost access” state.
- Scheduled scans stop within one cycle of failure detection.
- Change alerts and report emails stop.
- On request, scan data, evidence-store responses, and report PDFs are purged from object storage and the database within 30 days.
Customer data is never retained beyond the documented retention window for marketing, benchmarking, or analytics purposes after revocation.
3. Data handling
What we collect
The worker pulls configuration and posture data from your connected cloud APIs: user MFA or 2-Step Verification status, device compliance or ChromeOS inventory, directory admin role assignments, backup or retention policy summaries, and similar metadata.
We do not collect:
- Mailbox content, cloud file content, chat or meeting messages, calendar entries, or any other user-generated content.
- Authentication credentials, MFA secrets, recovery codes, or password material.
- Network telemetry, endpoint agent data, or device-side artefacts.
- Personal information beyond what is necessary to render a finding — a username on a “no MFA” line item, a device name on a “not compliant” line item.
Where it lives, and for how long
| Tier | What it holds | Retention |
|---|---|---|
scan_snapshots | Structured per-control signals derived from each scan, scored against ASD ML1/ML2/ML3 thresholds | Retained while the subscription is active (and for the free-assessment window on trials). Snapshot trim / summarise-after-N-days is not currently automated. |
| Evidence store (object storage) | Raw cloud API response bodies, hash-addressed by sha256(response) for the evidence chain | Plan-tiered: trial/Business ≈ 90 days; Growth ≈ 12 months; Enterprise unlimited / contract |
| Generated reports (object storage) | PDF assessments emailed to the customer | Retained for the active subscription; deleted on cancellation |
These responses contain security and configuration metadata only. Raw evidence bodies are posture data such as MFA registration status, device compliance state, and role assignments — not user content.
Shorter windows on request. Customers who need a shorter evidence retention window than their plan default can request one via [email protected].
Deletion on request. After you revoke access (admin consent removal or DWD deletion), you can additionally request full data deletion. We honour that within 30 days and confirm in writing.
What we never do with it
Hard commitments — reviewed each release and reflected in engineering practice:
- No machine-learning training.We never use customer scan data to train any machine-learning model — ours, a vendor’s, or a third-party’s. No exceptions.
- No third-party data sharing. Customer data is not sold, rented, or shared with brokers or advertising networks. The sub-processor list below is operational dependencies, not data partners.
- No cross-tenant correlation.Findings from one customer’s tenant are never combined with another customer’s data in the same query, report, or dashboard.
- No customer impersonation. Microsoft access uses the application service principal; Google access uses a delegated service account with DWD. We do not persist end-user OAuth tokens for scanning.
Encryption
- In transit. All cloud API traffic, all customer ↔ Aegis Eight traffic, and all internal service-to-service traffic is TLS 1.2+.
- At rest. The database, object storage, and secret store are encrypted at rest with managed keys.
Data residency
All customer data stays in Australia. Every scan snapshot, raw evidence body, generated report, and database row is stored and processed in our Australian hosting region (Sydney) by default.
- Storage and compute. Database, object storage, workers, and secrets run in Sydney.
- Email delivery. Report and alert emails are dispatched from within the Australian region.
- Limited exceptions. Some operational sub-processors may process contact email or authentication identifiers outside Australia. Cloud posture scan data is not among it.
Need a different region or cloud? Dedicated deployments in your required region are available on enterprise terms — contact us before any data is collected.
4. Tenant isolation
Customer tenants are isolated from each other by construction:
- Per-row scoping in the database. Every row carries a tenant key (
azure_tenant_id— Microsoft GUID or Google customer ID). All queries filter on it. - Per-tenant credentials.Microsoft scans use that tenant’s Entra ID in the OAuth flow; Google scans use that customer’s delegated admin and DWD scopes. There is no cross-tenant superuser token.
- Per-tenant evidence partitioning. Evidence objects are keyed under
evidence/{tenantId}/{scanId}/{responseHash}.json. - Single-tenant rendering paths. Dashboard and PDF code paths take a
tenantIdat every entry point — no shared cache that mixes tenants.
6. Sub-processors
The vendors that may see customer data in the course of delivering Aegis Eight:
| Vendor | Purpose | Data exposure |
|---|---|---|
| Cloud hosting provider | Compute, database, and object storage | All customer scan data and report artefacts |
| Microsoft (Microsoft Graph) | Source of Microsoft 365 scan data | The customer’s own M365 configuration data |
| Google (Workspace Admin & Chrome APIs) | Source of Google Workspace scan data | The customer’s own directory and policy metadata |
| Email delivery provider | Transactional email delivery | Customer email, organisation name, attached PDF in flight |
| Payment processor | Subscription billing | Billing identity only — no scan data |
| Authentication provider | Customer authentication for the dashboard | Aegis Eight dashboard logins; separate from the customer cloud tenant |
We will give 30 days’ notice via the dashboard and email before adding a new sub-processor that has access to scan data.
8. Compliance posture
Aegis Eight is designed against the principles of SOC 2 Type II, ISO/IEC 27001, and IRAP. Aegis Eight does not currently hold any of those certifications. We will not claim certification before it is true.
- For customers building toward an IRAP submission, reports are designed to supportthat submission — the evidence chain is built for IRAP-QAF — but the assessment itself is the customer’s, not Aegis Eight’s.
- We will sign appropriate non-certified contractual commitments (DPAs, security schedules, MNDAs) in lieu of certification where needed.
Indicative roadmap
| Milestone | Indicative timing |
|---|---|
| Independent penetration testing | Annually |
| SOC 2 readiness assessment | 2027 (target) |
| ISO/IEC 27001 evaluation | Future consideration |
For a security questionnaire, draft DPA, or detailed control-mapping, contact [email protected].
9. Frequently asked questions
10. Vulnerability reporting
Security researchers, customers, and the public are encouraged to report suspected vulnerabilities or data-handling issues to [email protected].
We acknowledge reports within two business days, prioritise within five, and disclose fixed issues to affected customers if customer data was demonstrably at risk.
11. Document history
| Date | Change |
|---|---|
| 2026-07-20 | Scan-flow caption: no configuration changes are ever made to your tenant. Access summary lists sourced from shared product.ts matrices. |
| 2026-06-20 | Cloud-neutral data handling; Microsoft 365 and Google Workspace platform picker for trust model, access, evidence chain, and revocation. |
| 2026-06-06 | Added at-a-glance access summary, scan-flow diagram, FAQ, retention scope clarification, and indicative compliance roadmap. |
| 2026-05-30 | Initial publication. Snapshot of trust commitments at MVP. |
Questions? [email protected] · Microsoft 365 permissions · Google Workspace permissions