Trust Centre

Available for all privileged roles including system owners, library admins, and cross-tenant commercial roles (partner, agent, consultant). Time-based one-time passwords with encrypted secret storage and single-use recovery codes.

Granular permission model with platform-privileged roles (system_owner, library_admin), cross-tenant privileged plans (partner, agent, consultant), and tenant-scoped standard users. Enforced on every API request.

Central privilege authority classifies all roles and plans. Partner, consultant, agent, and system_owner all operate under privileged-role governance with shortened JWT lifetime, MFA-capable architecture, tamper-evident audit logging, and elevated session monitoring.

Minimum 12 characters following NCSC three-random-words guidance. Optional Have-I-Been-Pwned breach checking.

Rate limiting on login (IP and email), registration, password reset, and shared content verification.

All privileged accounts (platform and cross-tenant) have 24-hour session lifetime (vs 7-day standard). Session version invalidation on password change.

Every privileged action is recorded with a SHA-256 hash chain. Each entry links cryptographically to its predecessor, making silent modification detectable.

System owners can view, filter, and search the complete audit trail directly in the admin interface.

Built-in hash chain verification utility. System owners can validate the integrity of the full audit log at any time.

Tokens, passwords, and API keys are automatically redacted from audit log entries before display.

Every LLM interaction is logged with prompt/response hashes, token counts, model identification, and moderation flags. Raw content is never stored by default.

Prompt injection detection and PII scrubbing on all inbound AI prompts. System prompts are isolated and non-user-editable.

10-category PII and safety scanner on all AI responses. Flags phone numbers, postcodes, card PANs, JS event handlers, and prompt injection markers.

Library content requires explicit human review acknowledgement before publication. Moderation flags must be reviewed by an administrator.

Only pre-approved AI models are used. Currently: GPT-4o via controlled API integration. No fine-tuning on user data.

Controls mapped against the OWASP LLM Top 10 risk categories. Documented in the AI Governance Framework.

All data queries are scoped by user_id, organisation_id, and client_org_id. Cross-tenant access is prevented at the query layer.

Partner, agent, and consultant accounts operate client workspaces with strict data boundaries. Client data is never visible to other tenants.

Roles that span multiple organisations (partner, agent, consultant) are classified as cross-tenant privileged with elevated monitoring, shorter sessions, and audit attribution.

Learner authentication uses a separate JWT namespace. Learner tokens cannot access admin or owner endpoints.

HSTS (preload-ready), X-Frame-Options DENY, X-Content-Type-Options nosniff, strict Referrer-Policy, Permissions-Policy, and Content-Security-Policy.

CSP deployed in report-only mode with violation reporting to an internal aggregation endpoint. Enforcement preparation underway.

Global 2 MB body size guard for JSON payloads. Upload routes enforce dedicated per-file limits.

TLS enforced via HSTS. Kubernetes ingress handles TLS termination with modern cipher suites.

All credentials loaded from environment variables. JWT secrets and encryption keys validated at startup. No hardcoded secrets in source.

Core technical controls aligned with Cyber Essentials requirements. Formal assessor submission and evidence review underway.

MFA enforcement, external penetration testing, and advanced session-security controls scheduled as part of the Cyber Essentials Plus readiness programme.

Security governance, audit logging, asset inventory, AI governance, and procurement evidence frameworks established as foundations for ISO 27001 readiness.

Cookie consent gating, data minimisation in AI prompts, tamper-evident audit logging, tenant isolation, and lawful basis documentation in place.

Limited-risk classification assessed. Transparency, human oversight, AI call logging, and moderation requirements documented and operational. Preparatory alignment with emerging EU AI Act principles.