Trust & Security
Lynctera is built for regulated environments. Security, data isolation, and AI governance are architectural properties of the platform — not policies added after deployment.
View Vanta Trust CenterLynctera is built for regulated environments. Security, data isolation, and AI governance are architectural properties of the platform — not policies added after deployment.
View Vanta Trust CenterWe invest in certifications that matter to regulated procurement teams. Request our SOC 2 report and review our policies via the Vanta trust center.
Independently audited. Report available on request via our trust center.
AI management system certification programme in progress.
Data processing agreements, consent management, and data subject rights built into the platform.
Each client operates within a segregated tenant environment, with dedicated data stores and tenant-specific security boundaries.
Every customer gets a dedicated Azure SQL database. No shared tables and no row-level filtering, so a query cannot reach another tenant’s data.
Tenant credentials, API keys, and certificates are stored in Azure Key Vault with per-tenant scoping. Accessed via managed identity only.
Knowledge base content is indexed per tenant, so retrieval for one organisation cannot return another’s content.
Customer data is stored in Microsoft Azure UK South, with backups in UK West. AI models run on our own Azure AI resource in Sweden Central (EU). For most models, Microsoft may process an individual request in another region where the model is deployed; data at rest stays in the UK and EU.
AI governance is enforced by the platform architecture, not by policy documents. These controls cannot be bypassed.
Every AI action is logged automatically: model used, tokens consumed, cost, duration, and output. LLM decisions and tool operations are tracked on separate audit surfaces.
Deterministic rules evaluate first. AI is invoked only when synthesis or judgement is required. This reduces cost, risk, and unnecessary model calls.
Workflows pause at defined gates until the person assigned approves. Anyone else is refused, and the refusal is logged. No AI agent can send email: replies are saved as drafts for a person to send.
The framework blocks AI calls with incomplete context. If required information is missing, the call is rejected before it reaches the model.
Agent prompts, models, and parameters are version-controlled with temporal history. Any AI behaviour can be reconstructed at any point in time.
AI configuration is continuously monitored across tenants and environments. Deviations from the governed baseline are detected and reported automatically.
Enterprise-grade infrastructure on Microsoft Azure. Production access is restricted to named engineers, behind multi-factor authentication and a certificate-based VPN, and database sign-ins are audited.
All services run on Microsoft Azure. Compute, storage, AI services, and networking are managed within Azure's compliance boundary.
No hardcoded credentials. Each application connects to its databases and to Azure Key Vault as its own Azure Managed Identity; any other secret is held in Key Vault and read at runtime.
Every AI model call goes through our own Azure AI resources, never a public model API: OpenAI models through Azure OpenAI, and Anthropic Claude models through Azure AI Foundry, for which Anthropic is a sub-processor. Customer data is not used to train models.
Public-facing services are protected by Azure Front Door with Web Application Firewall. Backend services have network-level access restrictions.
Visit our Vanta trust center to request the SOC 2 report, review our security policies, or ask a security question.