Data security & confidentiality

Your data stays yours.

SOPs, audit files, evidence and audit reports are confidential. Other clients and partners cannot reach them — that is enforced at database level. The AI vendor only receives what we send it, under commercial terms.

5 layers of protection

01

Database isolation

Row-Level Security in PostgreSQL — other clients can technically never reach your data, not even through faulty queries.

02

Server-side AI calls

The browser never talks to Claude directly. For each question, the server builds a request containing only your documents and context.

03

Invisible configuration

API keys and system prompts live on the server. Not in the browser, not copyable, not modifiable by users.

04

Prompt-injection protection

Transcripts, documents and auditor notes are fenced off in the prompt as data: instructions inside them are reported, not executed. The agent does not disclose its own system prompt, and other tenants are not in the context.

05

Anthropic under commercial terms

The platform runs on the Claude API under Anthropic's commercial terms: no training on your data, 30-day maximum retention, no Anthropic access to your stored documents. They reside in Ireland, the application in Frankfurt; the AI call itself goes to the global endpoint — Anthropic has no EU-only inference.

Commitments

What we guarantee
  • No access for other clients or partners
  • No training by Anthropic on your data
  • EU hosting (Supabase EU), TLS 1.3, bcrypt
  • Data processing agreement with every subprocessor — in force through incorporation into the terms of Anthropic, Supabase and Vercel, including the EU Standard Contractual Clauses (SCCs).
  • Right of access, rectification and erasure (GDPR) — part of the release; until then we handle such requests manually.

A check mark means it applies today. An arrow means it is scheduled for the release — we would rather say so up front than after the fact.

What is accessible
  • Your own employees — only their own data
  • Your quality manager — results within your own organisation
  • Your partner consultant — agent configuration, under NDA
  • RealAxces (support) — under NDA and GDPR
  • Anthropic — receives what we send, not used for training, retained for up to 30 days

A question from practice

"Could someone use Claude to pull our SOPs or audit reports out of your platform?"

No. Other clients' documents are never part of the AI request. Database isolation + server-side context building make cross-tenant extraction technically impossible.

Questions about security?

Ask them directly

Want to request the data processing agreements, walk through the architecture, or raise a specific audit question? An initial call is no-obligation and to the point.

FRANK@REALAXCES.NL