Security & compliance
Security and compliance, built in. Not bolted on.
Marque is built to pass enterprise security review. Customer briefs, brand identity, prompts and campaign artefacts remain customer-owned and are never used to train models. This page is the plain-language overview for your security, IT, procurement and legal teams. A fuller security pack — DPA, sub-processor list, controls walkthrough — is available on request.
Where we are
Status, plainly.
Marque is accepting applications. We use one of three honest states for every compliance claim. Built to means the controls are in place and operating — the formal audit is not yet complete. Held means the certificate exists and the report can be shared under NDA. In preparation means the document is being drafted.
| Standard / framework | Status | Evidence |
|---|---|---|
| SOC 2 Type II | Built to | Controls implemented; formal audit not yet pursued. Walkthrough on request. |
| ISO/IEC 27001 | Built to | Information Security Management System designed to ISO 27001 Annex A controls. Certification not yet pursued. |
| GDPR & UK GDPR | In preparation | DPA being drafted. Will be made available on request before general availability. |
| International transfers | In preparation | SCCs and UK IDTA addenda to ship with the DPA. |
Eight domains
The shape of the security posture.
SOC 2 · ISO 27001
Built to both. Annual audit cadence and report availability under NDA will follow formal certification. Penetration test programme planned before general availability. Status table above is the canonical reference.
UK · EU · US
Tenant-level pinning at provisioning time. Cross-region replication off by default — opt-in per contract. Current production environment runs in AWS eu-west-1 (Ireland). Schrems II addressed via SCCs and UK IDTA.
SAML · OIDC · SCIM
SSO from Okta, Azure AD, Google Workspace, Ping. Automatic provisioning and deprovisioning via SCIM. MFA enforceable per tenant. Role-based access control maps to the platform's role taxonomy with audit-logged role changes.
In transit · At rest · BYOK
TLS 1.2+ in transit, AES-256 at rest. Customer-managed keys via AWS KMS or Azure Key Vault available on the enterprise tier. Field-level encryption available for sensitive attributes. Automatic key rotation on platform-managed keys.
VPC · Private Link
AWS PrivateLink and Azure Private Endpoint for enterprise tenants. Dedicated tenant deployment available. IP allowlisting per tenant. No public ingress to the data plane of enterprise deployments.
Full audit log
Every checkpoint, prompt and approval is recorded with actor, timestamp and target. Real-time export to customer SIEM (Splunk, Sentinel) on the enterprise tier. Tamper-evident append-only storage. Retention configurable per tenant.
Model-agnostic · No training
Customer content is never used to train models. Marque routes to Anthropic, OpenAI, Google and self-hosted open models under zero-data-retention terms. Enterprise tenants pin per layer and per region; restrict to a single provider if required.
GDPR · UK GDPR
DPA with sub-processor list (in preparation). Right-to-erasure honoured down to the embedding layer. Data minimisation at ingestion; retention windows configurable per tenant. Breach notification within 72 hours per GDPR Article 33.
What we do with your data
Ingestion to deletion. In sequence.
-
01
Ingestion
Briefs, brand identity documents, voice fingerprints, reference assets and campaign inputs enter Marque via the application UI or signed API calls. Each artefact is tagged with its tenant on receipt — that tag determines every downstream access decision.
-
02
Processing
Content moves through the agent layers — strategy, concept, creative, production, activation, measurement. Each agent runs in a tenant-scoped context: no shared state with other tenants, no cross-tenant prompt or embedding bleed. LLM provider calls go out under zero-data-retention terms — your content is not retained by the provider beyond the request lifetime and is not used to train models.
-
03
Storage
Generated artefacts (concepts, copy variants, image / video / audio renders, audit trail) are written to the tenant's pinned region, encrypted at rest with AES-256. Rendered assets are watermarked. Cross-region replication is off by default.
-
04
Export & ownership
Brand identity, prompts and campaign artefacts are customer-owned. Export is available at any time via the application UI or signed API — full JSON for structured data, original-format files for media. No vendor lock-in by file format.
-
05
Deletion
On offboarding, customer data is deleted from primary stores within 30 days and from backups on the next backup rotation cycle. Right-to-erasure requests on individual records are honoured down to embedding tables. Audit logs are retained per the tenant's configured retention window before deletion.
Multi-tenancy
Tenant boundaries are not soft.
Every row in the database carries a tenant_id and every query in the application reads that id from the session context. Row-level-security policies in Postgres enforce the same rule at the database layer for the tables already migrated (2026-05-27 rollout, in progress per-table). Cross-tenant URL probes return no rows. Enterprise tenants run on dedicated infrastructure with their own database and storage.
Data, prompts, embeddings and context never cross a tenant boundary.
Sub-processors
Who else touches the data.
The canonical sub-processor list ships with the DPA and is currently in preparation. Customers will be notified of additions or changes in line with the standard 30-day DPA notice period. The active set today is summarised below.
| Sub-processor | Purpose | Region |
|---|---|---|
| Amazon Web Services | Hosting, storage, key management, LLM access via Bedrock | EU (Ireland) |
| Anthropic | LLM provider (text agents) — accessed via AWS Bedrock | EU (Ireland) via Bedrock |
| OpenAI | LLM provider (text + image generation) | US |
| Google AI Studio | LLM + image generation (Gemini, Imagen, Nano Banana) | US / global |
| Black Forest Labs, Runway, Kling, ByteDance, ElevenLabs | Generative image / video / audio providers (used only when the operator selects them per agent) | Provider-dependent |
Customer-pinned model routing on the enterprise tier lets you restrict generation to a single provider or region, removing all others from the active set for your tenant.
For your security review
The questions enterprise teams actually ask.
Do you use our data to train AI models?
No. Customer briefs, brand identity, prompts and campaign artefacts are never used to train models. Marque routes requests to third-party LLM providers (Anthropic, OpenAI, Google) under their zero-data-retention terms; customer content is processed only for the duration of the request.
Where is our data stored, and can we pin a region?
Tenants are pinned at provisioning time to UK, EU or US data residency. Cross-region replication is off by default. The current production environment runs on AWS eu-west-1 (Ireland). For corporate customers requiring other regions or dedicated tenancy, this is set up per contract.
Can we use our own encryption keys?
Bring-Your-Own-Key (BYOK) via AWS KMS or Azure Key Vault is available on the enterprise tier. Default tenants use platform-managed AES-256 at rest with automatic key rotation.
Can Marque be deployed in our VPC?
Yes, on the enterprise tier. We support AWS PrivateLink and Azure Private Endpoint with dedicated tenant deployment. The infrastructure-as-code is handed over for your security team to review.
Which LLM providers do you use, and can we restrict to one?
Marque is model-agnostic. Today the platform is wired to Anthropic (via AWS Bedrock), OpenAI, Google AI Studio, Black Forest Labs (Flux), ElevenLabs, Runway, Kling and ByteDance Seedance. Enterprise customers can pin to a specific provider, region or self-hosted model, and route different agent layers to different models.
How is one client's data isolated from another's?
Every row in the database is scoped by tenant_id, enforced at the application layer through a session-derived tenant context. Postgres row-level-security policies are being layered on per table during the 2026-05-27 isolation rollout. Data, prompts, embeddings and context do not cross a tenant boundary.
What certifications do you hold, and can we see the reports?
Marque is accepting applications. The platform is built to SOC 2 Type II and ISO/IEC 27001 controls (encryption, access management, audit, change control, vendor management, business continuity) but is not yet formally audited. A DPA covering GDPR and UK GDPR is in preparation and will be available for review on request.
What happens to our data if we leave?
Brand identity, prompts and campaign artefacts are customer-owned and exportable at any time. On offboarding, customer data is deleted from primary stores within 30 days and from backups on the next backup rotation cycle. Right-to-erasure requests are honoured down to the embedding layer.
How do you handle a data breach, and what's your notification timeframe?
On confirmed breach affecting customer data, affected customers are notified as soon as practicable and within 72 hours, aligned with GDPR Article 33. The notification includes scope, affected categories of data, remediation actions taken and contact information for the response team.
Do you have a vulnerability disclosure programme?
Yes. Security researchers can reach the team through the form on this page. Reports are acknowledged within one working day. A formal bug bounty programme will be added before general availability.
Get in touch
Bring your questionnaire. We've answered it before.
Whether you need the DPA, want to share a SIG or CAIQ, are asking about a specific control, or are reporting a vulnerability — this form reaches the team directly. Acknowledgement within one working day.