Trust
Security and data processing
Meritfold ingests the documents a company uses to win work: past bids, policies, certifications and case studies. This page describes how that material is handled, and what we deliberately do not claim.
Last updated 31 August 2026
Controls that are properties of the system
Isolation enforced by the database
Every tenant table has PostgreSQL row-level security enabled and forced, and the application's own database role cannot bypass it. With no tenant context set, nothing is readable or writable at all. This is a property of the schema, not a filter the application remembers to apply.
Your documents are not training data
Analysis and drafting call a model provider's API. Under those API terms, content sent to the API is not used to train their models. Meritfold itself trains no models on your content, and nothing you upload can reach another organisation's results.
Untrusted text stays untrusted
Text from your documents and from a buyer's tender pack is always carried as data, never as instructions. A document that attempts prompt injection raises a security event for the operator rather than being obeyed.
Claims are verified after the model answers
An extracted claim whose verbatim quote cannot be located in its source page is discarded, and a response carrying an unsupported claim cannot be exported. That is enforced in code after generation, not requested of the model beforehand.
An approval trail that cannot be edited
Approvals, audit events and outcomes are append-only: the application holds insert and select rights on those tables and nothing more. Each approval records who approved, when, and a hash of exactly what they approved.
Leaving is a supported operation
A full export of every table plus your original documents and generated artifacts, which an owner or admin runs from Settings without asking us. Permanent deletion removes rows, vectors, stored objects and artifacts, and runs as the same kind of audited workflow, but we trigger it on your instruction rather than exposing it as a button, so an account cannot be destroyed by a single click.
Where your data is
Customer data is stored on dedicated infrastructure in Germany (EU). Traffic is served over TLS. Database backups run nightly with restricted file permissions, and restoring from them has been tested rather than assumed. During the founding-partner period the application is reachable only over a private network, not the public internet.
Subprocessors
We use a deliberately short list, and we will tell you before it changes.
| Subprocessor | Purpose | Data involved |
|---|---|---|
| OpenAI | Document analysis, scoring and drafting | Excerpts of the documents a task needs |
| Hetzner | Server hosting and storage | All application data at rest |
What we do not claim
A security page that only lists strengths is not much use to a buyer. These are the gaps as they stand today.
- We do not hold SOC 2, ISO 27001 or Cyber Essentials certification today. If a certification matters to your procurement process, tell us and we will tell you honestly where we are.
- We have not commissioned a third-party penetration test yet. Our own security review and its findings are documented, and we will share that during onboarding.
- We do not operate a 24/7 security operations centre. Meritfold is early-stage software run by a small team.
- We do not offer a contractual uptime guarantee during the founding-partner period.
Reporting a vulnerability
We do not publish a security mailbox yet. Until we do, report an issue through the person who onboarded your organisation, or by post to FusionX Media Ltd, 4th Floor, Silverstream House, 45 Fitzroy Street, Fitzrovia, London W1T 6EB, United Kingdom. We will not pursue anyone who reports a genuine issue in good faith.