Trust architecture

Say the control that exists — and the boundary it actually protects.

Benedam is sold to technical security teams. Our public claims are deliberately narrower than our ambitions and change only when a control is deployed, tested, and evidenced.

Deployed design

Application and tenant security

These statements describe implemented architecture, not certifications or future controls.

Database-per-tenant

Each tenant has a separate PostgreSQL database and role. This is a strong defence against application-layer cross-tenant mistakes; it is not protection from a fully compromised host.

Row-level security

PostgreSQL RLS is applied inside the tenant database as defence in depth. Tenant identity is set with SET LOCAL inside a transaction so pooled connections cannot retain another caller’s context.

Per-host sessions

Cookies use the __Host- prefix and never Domain=.benedam.com. Tenant access tokens are bound to the host audience; cross-host transitions use a short-lived handoff code.

AI permission ceiling

ROSS has no database network access and no write tools. It forwards the caller’s token to read-only domain APIs, and tenant identity never comes from a model-supplied argument.

Operator validation

Automated scan results are drafts. Only an authorised BalticSec operator can validate and publish a finding to a client.

Engine isolation

The pilot engine uses a dedicated Redis stream identity and scoped callback token. It has no PostgreSQL route, no Docker socket, and no privileged container access.

Data and infrastructure

Current posture and explicit readiness gates

EU hosting is active. Some customer-readiness controls remain blockers and are not marketed as complete.

Implemented
  • EU-hosted application and control plane
  • Cloudflare proxy with strict origin TLS
  • Service memory and log limits
  • Immutable application image deployment model
Customer scanning gate
  • Encrypted customer engine storage
  • Observed offsite backup timers
  • Scratch tenant and evidence restore drill
  • Verified customer engine end-to-end run

Real customer scanning and evidence retention remain blocked until every gate is completed and recorded.

Disclosure

Found a security issue?

Please use our coordinated disclosure process. Do not test production customer workspaces or access data that is not yours.

Responsible disclosure policy
Start with evidence

Turn the next security test into a continuous assurance programme.

Explore a read-only product tour, or tell us what you need to prove and we will scope it with you.