SOC 2 status
The SOC 2 Type II control map to the running system, and where the observation window stands.
- Version
- 1.0.0
- Passing
- 13
- Total
- 17
- Ai inventory problems
Observation window
- Opens at gate
- G2
- Status
- not_opened
- Auditor
- DEMO-Audit LLP (fictional)
Controls
| Id | Tsc | Title | Mechanism | Evidence | Owner | Frequency | Evidence ok |
|---|---|---|---|---|---|---|---|
| CC1.4 | Control environment | Engineering guardrails are documented and enforced | CLAUDE.md hard rules R1-R14; CI gates per PR | ci_gates | DevOps | per_change | no |
| CC2.2 | Communication | Sub-processors and AI systems are disclosed | trust centre generated from the system inventory | trust_centre | Compliance | per_deploy | yes |
| CC3.2 | Risk assessment | Jurisdiction risk is evaluated before surfacing restricted offerings | jurisdictionGate on every path (R12) | jurisdiction_gate | Compliance | continuous | yes |
| CC5.2 | Control activities | Policy-as-code for agent access and eligibility | Cedar policy packs under /policies | policy_artefacts | Security | per_change | yes |
| CC6.1 | Logical access | Tenant isolation on every record and route | tenant_id on every record; route-level tenant check (R8) | tenant_isolation | Security | continuous | yes |
| CC6.6 | Logical access | Inbound agent traffic passes the trust edge | trust-edge middleware on protocol routes (R3) | trust_edge | Security | continuous | no |
| CC6.7 | Logical access | Sensitive data classes are encrypted with per-tenant keys | AES-256-GCM, HKDF per tenant and data class (ADR-016) | tenant_keys | Security | continuous | yes |
| CC6.8 | Logical access | Malicious code and vulnerable patterns are blocked | Semgrep SAST, secret and PAN scans in CI (ADR-015) | sast | Security | per_change | no |
| CC7.2 | System operations | Security events are logged immutably | hash-chained audit log, per-agent-action audit with 5-year retention | audit_chain | Security | continuous | yes |
| CC7.3 | System operations | Evidence chain integrity is verified on a schedule | nightly evidence verifier | evidence_verifier | Security | daily | yes |
| CC8.1 | Change management | Changes pass review and CI before deploy | branch protection, required checks, event-schema compatibility gate | ci_gates | DevOps | per_change | no |
| C1.1 | Confidentiality | Confidential classes never reach logs, captures or prompts | scrubForModel + span scrubber; PHI scrubber tests in CI | scrubber | Security | per_change | yes |
| C1.2 | Confidentiality | Retention schedules are enforced | retention schedules per data class; purge job | retention | Compliance | daily | yes |
| P4.2 | Privacy | Consent per channel with withdrawal evidence | consent records per channel; propagation <= 60 s | consent | Compliance | continuous | yes |
| P5.1 | Privacy | Data-subject requests across services | DSR export/delete across every store; drill | dsr | Compliance | monthly | yes |
| PI1.3 | Processing integrity | Generated content carries AI marking | marking manifest verified before publish (R11) | marking | AI engineering | per_change | yes |
| A1.2 | Availability | Data residency is enforced per region | RegionalStore residency guard; PHI and payment-adjacent pinned without exception | residency | DevOps | continuous | yes |
Iso42001
- Standard
- ISO/IEC 42001:2023 (AI management system)
- Assessed at
- 2026-10-26T00:00:00.000Z
- Owner
- AI governance lead (DEMO)
Items
| Clause | Title | Status | Gap | Action | Owner | Due gate |
|---|---|---|---|---|---|---|
| 4.3 | Scope of the AI management system | partial | Scope statement not yet approved by the owner | Owner signs the AIMS scope covering the inventory systems | Compliance | G2 |
| 5.2 | AI policy | gap | No published AI policy | Publish AI policy in the trust centre | Compliance | G2 |
| 6.1.2 | AI risk assessment | partial | Risk register exists per system but without likelihood/impact scoring | Score each inventory system; review quarterly | AI engineering | G3 |
| 6.1.4 | AI system impact assessment | gap | No impact assessment for branded agents and chat surfaces | Impact assessment for systems that interact with natural persons | AI engineering | G3 |
| 7.5 | Documented information | met | none | AI-system inventory generated from the system and exported from the trust centre | Compliance | G2 |
| 8.2 | AI risk treatment | partial | Treatments exist (disclosure, marking, eval sets) but are not linked to risks | Link controls to risks in the register | AI engineering | G3 |
| 8.4 | AI system life cycle | partial | Evaluation sets per task (LP-03-03) not yet complete for every task | Eval set per model task; below-threshold fails CI (gate 6) | AI engineering | G2 |
| 9.1 | Monitoring, measurement, analysis | partial | Budget and usage visible per tenant; model quality drift not yet monitored | Drift dashboards per task | AI engineering | G3 |
| A.6.2.8 | Event logs of AI systems | met | none | ai.disclosure.recorded and per-agent-action audit with 5-year retention | Security | G2 |
| A.8.2 | Information for users (transparency) | met | none | R11 disclosure text injected by runners; marking manifests on generated content | AI engineering | G2 |
| A.10.3 | Suppliers (model providers) | gap | No model-provider contracts reviewed (no provider keys in the demo) | Supplier review for each llm-gateway provider before G2 | Compliance | G2 |
Summary
Met: 3 Partial: 5 Gap: 3Source: compliance/api.ts soc2Status()