SCIFAI — Sovereign AI

The AI room cleared for sensitive work.

A Sensitive Compartmented Information Facility doesn’t prohibit sensitive conversations — it creates the certified environment in which they can happen freely, safely, and without risk of compromise. SCIFAI does the same for artificial intelligence: a sovereign, accreditation-ready environment where your organisation can use AI on client files, financial models, and privileged work, because data never leaves the perimeter and every action is attributed.

  • Zero data egress — blocked at the network layer
  • 100% audit coverage with immutable attribution
  • Accreditation-ready: ISO 27001, SOC 2, CMMC 2.0
  • NATO-aligned, NRC-IRAP funded (Project 1041303)

Trust is not a feature. It is the architecture.

Every design decision answers one question: what would need to be true for a law firm partner or a bank’s chief risk officer to use AI on their most sensitive work without hesitation?

Sovereignty

Your data is processed exclusively within your controlled environment — never routed through third-party infrastructure, never used to train external models, never subject to a model provider’s terms of service. You own the environment and the data.

Attribution

Every AI interaction is tied to a verified identity. Every prompt, completion, and data access is recorded. There is no anonymous AI usage — when a regulator or client asks who used AI and on what, the answer is immediate.

Containment

Data classified as sensitive cannot leave the perimeter. Egress is blocked at the network layer. There is no mechanism by which a user — intentionally or accidentally — can push sensitive data to an external system.

Accreditation

SCIFAI is designed to be audited. The architecture, controls, and audit logs are structured to support accreditation against ISO 27001, SOC 2, CMMC 2.0, and sector-specific regulatory requirements.

Five layers. One perimeter.

Each layer is independently auditable. A failure in any single layer does not cascade — the same defence-in-depth principle that governs physical SCIF construction.

01

Perimeter Layer

Network isolation & ingress control

All traffic enters through a single, monitored ingress point. No direct internet access. Egress is blocked by default. Every connection is logged, inspected, and attributed to an authenticated identity before any AI processing begins.

Zero Trust NetworkEgress BlockingTLS Inspection
02

Identity Layer

Authentication & authorisation

Multi-factor authentication is mandatory. Role-based access control determines which AI capabilities each user can invoke and on which data classifications. No anonymous AI usage — every session is tied to a verified identity.

MFA EnforcementRBACSession Attribution
03

Data Layer

Classification, containment & encryption

Data is classified at ingestion. Sensitive classifications are processed in isolated compute environments. Encryption at rest and in transit is enforced. Data never leaves the sovereign perimeter — no training, no telemetry, no third-party routing.

Data ClassificationSovereign ComputeAES-256 Encryption
04

AI Model Layer

Controlled inference & model governance

AI models run within the perimeter. Model versions are pinned and audited. Prompts and completions are logged with full attribution. Model behaviour is governed by organisational policy, not by the provider’s terms of service.

On-Premise InferencePrompt LoggingModel Pinning
05

Audit Layer

Attribution, compliance & reporting

Every AI interaction — who asked, what was asked, what was returned, what data was accessed — is recorded in an immutable audit log. Reports align with regulatory requirements, and compliance evidence is available on demand.

Immutable Audit LogRegulatory ReportingCompliance Evidence

Built for organisations where exposure is not an option.

Where banning AI is a competitive disadvantage and routing data through third-party tools is a breach of duty, SCIFAI is the certified alternative.

Legal & Professional Services

Privilege-preserving AI for law firms

Solicitor-client privilege is a legal duty, not a policy. SCIFAI lets legal teams work with client data freely — contract review, due diligence, legal research — with a full audit log and zero data egress.

SRA Code of ConductUK GDPRLaw Society AI Guidance

Financial Services

Accountable AI for regulated institutions

FCA Consumer Duty and OSFI B-13 require firms to demonstrate how AI is used in client-facing decisions. SCIFAI provides the audit trail, attribution, and containment regulators require and risk committees can verify.

FCA Consumer DutyOSFI B-13MiFID IIBasel III

Defence Supply Chain

Compliant AI for sensitive projects

Defence contracts increasingly contain AI prohibition clauses because no certified alternative exists. SCIFAI is that alternative — a sovereign environment that satisfies prime contractors and government clients so SME suppliers can use AI without breaching contract.

CMMC 2.0NATO CSDMITAR / EARISO 27001

The numbers that matter inside the perimeter

0

Data egress events in any SCIFAI deployment

100%

Audit coverage of every AI interaction

<30

Days from briefing to production deployment

The conversation starts inside the perimeter.

An executive briefing is a private, 60-minute session with the SCIFAI team — the architecture, the deployment model, and the regulatory obligations specific to your organisation, conducted under NDA.