Security Model
Kilimanjaro's security model decomposes into five threat-model surfaces — five concentric layers from the innermost hardware perimeter to the outermost query layer — plus the audit ledger that captures every layer's contribution. Every surface enumerates the threat classes it defends against and the structural mitigations the platform applies. Compliance composes against the structural posture; it does not substitute for it.
1. The five threat-model surfaces
From innermost to outermost:
| Surface | What it defends | Substance |
|---|---|---|
| Infrastructure | The hardware and hypervisor perimeter | Hardened baseline substrate with mutual TLS + SPIFFE workload identity; Confidential Computing (TDX / SEV-SNP) as a Secure Key Release (SKR) mode overlay; hypervisor-level intrusion mitigation; the limits of TEE protection |
| Cryptographic Boundary | Key lifecycle and transmission | Per-project software-protected KEK in a region-pinned vault at baseline, rooted in a regional FIPS 140-3 Level 3 HSM-backed master key; plaintext sequestration in TEE memory in SKR mode; the DEK / CEK separation; transmission vectors; key rotation discipline |
| Isolation & Privilege | Project boundary and privilege compartmentalisation | Three independent cross-project isolation layers (DB credential scoping, query-planner scoping, connection-pool routing); privilege compartmentalisation per project (six role-scoped database users; the death of the root key) |
| Identity-Aware Kernel | Session and token integrity | Platform-signed workload identity at baseline; hardware-attested workload identity and HSM-backed token signing in SKR mode; defending against forged JWTs, impersonation, persistent credential stuffing |
| Injection | Query-layer manipulation | The four-layer anti-injection posture (SDK → Edge → Origin → DB); the rejection log; STRIDE classes the surfaces close |
The surfaces are concentric, not parallel. A request entering the platform passes through Injection-layer pattern matching at the edge, hits Identity-Aware Kernel for token verification, traverses Isolation & Privilege boundaries as it accesses project resources, exercises Cryptographic Boundary if the request requires unwrapping a key, and runs inside Infrastructure-layer TEE protection throughout. Each layer enforces independently; failure of any one layer does not silently fall through to the next.
2. The hardened execution substrate
All five surfaces share a hardened execution substrate. Every platform workload runs on a mesh-secured, hardened foundation: service-to-service communication is mutual TLS with SPIFFE workload identity; per-project keys are software-protected secrets in region-pinned vaults; DEKs are used in-process and zeroized after use.
In SKR mode, this substrate is extended with a hardware isolation layer: workloads run inside a Confidential Computing TEE (Intel TDX on GCP / Azure, AMD SEV-SNP on AWS / OVHcloud / Hetzner), and attestation allows customers to verify — via the silicon vendor's attestation service — that any given platform component is running the expected code in the expected hardware configuration.
The Infrastructure surface documents both the hardened baseline posture and the SKR mode extension — its additional protections, its limits, and what a compromise at this layer would mean for the others.
3. The audit ledger as the forensic substrate
Every security layer emits audit events recorded in the platform's tamper-evident, hash-chained audit ledger, signed with the platform's RSA-2048 + SHA-256 key. The audit chain is anchored against an external transparency log, so customer compliance evidence packages can attest tamper-evidence without trusting the platform's word for it.
The combination — concentric defence-in-depth surfaces, a hardened execution substrate, and an externally-anchored audit chain — is the structural security posture. Compliance frameworks (SOC 2, ISO 27001, GDPR, HIPAA, FedRAMP) compose against this posture; they do not substitute for it.
4. How Compliance composes
Compliance documents the regulatory and industry frameworks Kilimanjaro pursues. The structural posture reduces the surface area a compliance attestation needs to cover — the platform's controls already enforce many framework requirements (data residency, access control, audit retention, encryption at rest and in transit), so attestation cycles document evidence of the existing controls rather than implementing controls from scratch per framework.
The shared-responsibility split between platform and customer varies by framework; the platform-side scope is the structural posture documented across these threat-model surfaces; the customer-side scope is application-layer controls (the customer's own RBAC role authoring, their own data-subject-rights workflows, their own incident-response procedures composed against the platform's audit log).
5. Responsible disclosure
The platform operates a responsible disclosure programme — scope, contact, response timelines, safe-harbour terms, and recognition for researchers who surface vulnerabilities. Reports inside the disclosed scope receive coordinated response; reports outside scope (third-party services, customer-managed KMS endpoints, customer SIEM endpoints) are routed to the relevant vendor.
6. See also
- Architecture overview — the platform's architectural shape at a glance.
- Identity-Aware Kernel — Gate as the structural identity decision.
- Custody architecture — the four-mode custody ladder and the upgrade/downgrade asymmetry.
- Threat Model — the five concentric surfaces in detail.
- Compliance — compliance frameworks + attestation surface.
- Responsible Disclosure — the vulnerability reporting programme.