00Posture at a glance
ModelCop is built to a secure-by-design, prove-every-claim standard. Isolation boundaries, alerting, and recovery procedures are verified — not just diagrammed.
01Tenant isolation
ModelCop is multi-tenant at the application layer but enforces tenant boundaries at the database layer, with multiple independent mechanisms so that no single application bug can cross a tenant boundary.
Application
Tenant context is strict and request-scoped. Any code path that reaches the database without an active tenant raises an explicit error rather than defaulting to one.
Database role
The app connects under a dedicated, least-privilege role that cannot bypass row-level security. The administrative role is used only for migrations — never to serve application traffic.
Database engine
Every tenant-scoped table enforces PostgreSQL Row-Level Security with force enabled, so the engine itself refuses rows outside the active tenant.
02Defense in depth
| Layer | Controls |
|---|---|
| Network | VPC-isolated; application and database in private subnets with no public IPs. A single load balancer is internet-facing, behind a managed web application firewall (OWASP, known-bad-inputs, and SQLi rule sets). |
| Transport | TLS 1.2/1.3 only; older protocols rejected. HTTP is redirected to HTTPS; plaintext is never served. Certificates auto-renew. |
| Application | OIDC single sign-on (no password storage), hardened session cookies (HttpOnly, Secure, SameSite), and secrets held in a managed, KMS-backed store — never in code. |
03Data protection
- At rest: database storage encrypted with AES-256; object storage encrypted with public access blocked at the account level.
- In transit: HTTPS externally and encrypted, certificate-verified connections to the database.
- Backups: continuous point-in-time recovery with 7-day retention; deletion protection enabled on the database cluster.
- Secrets: credentials and signing keys live in a managed secret store, read at runtime via scoped IAM roles, and are rotatable.
04Monitoring, detection & response
Management-plane activity, network flow logs, database logs, and application logs are captured to dedicated, access-controlled stores. Continuous threat detection is enabled, and real-time rules route critical events — root-account use, IAM and security-group changes, threat-detection findings, and failed sign-ins — for review. Image vulnerability scanning runs on every build.
05Disaster recovery
Recovery is validated end-to-end, not merely configured. A point-in-time restore exercise — new cluster, schema and role verification, tenant-isolation check, and teardown — was completed in May 2026 with all verification criteria passing, achieving the objectives below. We re-validate on a recurring schedule and after major infrastructure changes.
06Vulnerability management
Container images are scanned on every push; the current image carries no critical, high, medium, or low findings. Dependencies are pinned and reviewed before merge. Infrastructure is managed as code with state reconciled against reality. A third-party penetration test is planned post-launch.
07Compliance posture
We are explicit about what is in place today versus on the roadmap.
| Item | Status | Detail |
|---|---|---|
| Encryption & isolation | Live | AES-256 at rest, TLS in transit, 3-layer tenant isolation |
| Audit & threat detection | Live | Management-plane audit log, flow logs, threat detection |
| SOC 2 Type II | In progress | Readiness underway; target Q4 2026 |
| ISO/IEC 42001 | In observation | AI-management-system controls implemented |
| HIPAA-eligible architecture | Available on enterprise | BAA available; architecture structured for PHI workloads |
| GDPR data-subject rights | Live | Per-tenant export and deletion |
| Third-party pen-test report | Planned | Scheduled post-launch |
| FedRAMP | Not planned | Out of scope at this stage |
Under the cloud shared-responsibility model, ModelCop owns everything above the hypervisor; the infrastructure provider secures the underlying platform. The full security overview and questionnaire responses are available to prospects under NDA.
08Responsible disclosure
Found something? Tell us.
We welcome reports from security researchers. Act in good faith, don't access data that isn't yours or degrade the service, and we'll treat your research as authorized. We acknowledge reports within 24 hours and update every 7 days until resolution.