# AEGIS — Security Assurance Roadmap

- **Document ID:** ARCH-57
- **Phase:** E.4 — Readiness & Long-View
- **Status:** Draft for review (post four-reviewer discipline)
- **Version:** 0.1
- **Date:** 2026-07-10
- **Owner:** Chief Security Architect

---

## 1. Purpose

Set the path from the v0-GA assurance posture (ARCH-27 §3) to Level-5 targets across every critical subsystem. Ties together verification instruments, formal-methods pilots (ARCH-40), and Debt Register retirements (ARCH-29).

## 2. Roadmap by Subsystem

### 2.1 Security Kernel
- **v0-GA (L3).** Property tests (Golden Invariants) + Watchdog quorum + RIV + Independent audit.
- **v0.2 (L3→L4).** Expanded property tests + expanded chaos scenarios.
- **v1 (L4).** TLA+ model of Capability Directory + Kernel Decision Loop.
- **v1.5–v2 (L4→L5).** External formal-methods review + threshold cryptography pilot + Confidential Computing pilot.
- **v3 (L5).** All above + long-term sustained metrics.

### 2.2 Cryptographic Identity
- **v0-GA (L3).** SPIFFE-shape SVIDs + short-lived leaves + rotation ceremonies.
- **v1 (L4).** Hybrid PQC signatures piloted on long-lived audit archival.
- **v2 (L4→L5).** PQC primary across all signature sites.

### 2.3 Audit Chain
- **v0-GA (L4).** Layer-A hash chain + Watchdog Audit-in-audit + verifier CLI.
- **v0.2 (L4).** Transparency log (Sigstore Rekor + self-hosted witness) integrated.
- **v1 (L4→L5).** Formal analysis of monotonicity invariants.

### 2.4 AI Safety Layer
- **v0-GA (L3).** Deterministic validators + Golden Invariants + red-team engagements.
- **v0.2 (L4).** Multi-model consensus refined + canary corpus grown + purple-team runs.
- **v1 (L4).** Continuous adversarial testing + Adaptive Trust integrated.
- **v2 (L4→L5).** Formal reasoning about specific invariants (redaction correctness, no-conclusion-from-AI).

### 2.5 Evidence Engine
- **v0-GA (L3).** Signed provenance + dual-store ledger + reproduction canaries.
- **v1 (L4).** Formal reasoning about transformation-ledger consistency.
- **v2 (L4→L5).** Attested evidence exports; TEE-based attestations.

### 2.6 Recovery Engine
- **v0-GA (L3).** Recovery Integrity Attestation + drilled runbooks.
- **v0.2 (L3→L4).** Ransomware simulation + air-gap drills quarterly.
- **v1.5 (L4).** Formal specification of RIA invariants.

### 2.7 Plugin Isolation
- **v0-GA (L3, interface-only).** Wasm sandbox pattern defined; runtime v0.2.
- **v0.2 (L4).** Full runtime + isolation testing.
- **v1 (L4).** Publisher tier maturity + Adaptive Trust integrated.

### 2.8 Supply Chain
- **v0-GA (L3).** SLSA-3+, two-CI reproducibility, transparency log for release hashes.
- **v0.2 (L4).** Model provenance canary evals baseline.
- **v1 (L4).** Threshold cryptography for signing sub-roots piloted.

### 2.9 Other Engines
- Detection, Case, Connector, Normalization, Storage, Telemetry, Policy, Risk, Response — v0-GA L2/L3; v1 L3; v2 L4 where criticality warrants.

## 3. Milestones

- **v0-GA (year 1).** Achieve L3 baseline + eliminate v0-GA Debt Register entries per targets in ARCH-29 §4.
- **v0.2 (year 1 + 3 months).** Plugin runtime; expanded canary + red-team.
- **v1 (year 2).** Dashboard/Config Review/Asset Inventory/Risk Scoring/Threat-Intel; L4 targets for most; PQC + SAML + BYOK.
- **v1.5 (year 3).** Response Engine narrow; Endpoint agent limited GA; TLA+ delivered on Kernel.
- **v2 (year 4).** MSSP hosted; PQC primary; Confidential Computing pilots.
- **v3 (year 5+).** L5 for all critical subsystems; FedRAMP Moderate; broad Confidential AI.

## 4. Blockers and Dependencies

- **Formal-methods team ramp.** Blocks TLA+ delivery on Kernel; on Q-06-3 pilot.
- **PQC library maturity.** Blocks hybrid signature adoption timing.
- **TEE hardware availability.** Blocks Confidential Computing pilots.
- **Team scale.** Blocks MSSP hosted GA + expanded assurance work.

## 5. Assumption (hypothesis)

- **H-1.** *L5 achievable for critical subsystems within 5 years given sustained investment.*
   - Evidence: industry parallels; internal team velocity.
   - Validation: milestone-by-milestone retrospective.
   - Confidence: Medium.
   - Expiration: annually.
   - Review Date: 12 months.

## 6. Trust Score Contribution

Roadmap progress feeds `Verification` dimension over time.

## 7. Independent Architecture Review

- **F-1.** *Roadmap is a plan; plans slip.* Milestone-by-milestone re-planning.
- **F-2.** *Assurance targets may be aspirational.* Explicit in table + Debt Register.

## 8. Adversarial Architect Review

- **A-1.** *Attacker exploits gap between claimed and actual assurance.* Debt Register + Trust Score + auditor visibility.
- **A-2.** *Roadmap edited to hide slip.* Ceremony-controlled (ARCH-22 catalog class).

## 9. Operational Reliability Review

- **O-1.** Governance council reviews quarterly.
- **O-2.** Adjustments recorded via ADR.

## 10. Self-Critique

- **S-1.** *Dates could ossify.* Explicit re-planning cycle.
- **S-2.** *"L5" is aspirational.* Documented; incremental progression celebrated.

## 11. First-Target Analysis and Redesign

**Target:** the "declared" assurance level. If we claim L4 without evidence, everything downstream is false confidence. Response: assurance level is evidence-gated per ARCH-43; drift monitor; auditor annual audit.

## 12. Future Risks / Known Limitations / Out-of-Scope / Retirement

- **Future Risks.** Technology or attacker landscape shifts may require re-scoping.
- **Known Limitations.** Roadmap is directional; specific quarters flex.
- **Out-of-Scope.** Assurance for retired subsystems.
- **Retirement Conditions.** Continually renewed.

## 13. Decisions

### D-57-1. Per-subsystem assurance roadmap with milestone-anchored targets

- **Reason.** Mandate #36 + explicit long-view.

## 14. Change Log

- **0.1 (2026-07-10)** — Initial draft.