Security and trust center
Built to be examined.
FraudBrain handles evidence about people and money. This page describes how we treat that responsibility: what comes in, what never does, how long it stays, who can see it, and who to call. The full documentation set, including the architecture detail this page deliberately omits, is available during vendor review.
Posture
Six commitments. Examined on request.
Hosted on Google Cloud
FraudBrain runs on Google Cloud infrastructure. Encryption in transit and at rest is the default posture, not an option someone remembered to enable.
Scoped per organization
Every request is evaluated against the caller's organization on the server. Each organization sees exactly its own book. The boundary is proven by isolation tests that block deploys when they fail.
Invite-scoped access
Sign-in uses one-time codes delivered to pre-loaded invites. Access exists because an administrator granted it, and it is revocable the same way.
Least data
We take the fields the graph needs and no more. The metrics on this website are aggregate counts only; nothing on the marketing surface touches record-level data.
Human accountability
Reviewer actions are logged: who looked, who decided, when and why. The audit trail is part of the product, not an afterthought.
Documentation on request
Security and data-handling documentation is available during vendor review. Bring your diligence process and your data people.
Data handling
What comes in. What never does.
The product runs on business records a customer supplies under its data-access agreement, and on nothing else. The agreement names the fields before the first record moves.
The records the graph is built from
- Applications and quotes, with the producer, applicant and instrument fields the graph joins on.
- Payment and commission ledger records, so the exposure side of every producer is real, not estimated.
- Post-sale carrier evidence: what happened to a policy after issue.
- Hierarchy relationships: who recruited whom, and where volume actually comes from.
- The e-signature and origin telemetry attached to those records.
What stays out
- Anything about visitors to this website. Visiting fraudbrain.ai places nobody in a scoring system, and website analytics never reach the product.
- Any record outside the agreement. If a field is not named, it is not taken, and it cannot be added by a click.
- Any organization's data in another organization's view. Tenancy is enforced on the server, not styled in the interface.
- Record-level data on this marketing surface. The numbers here are aggregate counts only.
FraudBrain is advisory only. A score determines no action; people decide.
Controls
Retention, encryption, and who can see what.
The agreement sets the clock
Retention follows each customer's data-access agreement, not a default we picked. The specifics travel with the vendor-review packet.
In transit and at rest, by default
Every connection is encrypted in transit and every store is encrypted at rest on Google Cloud. Secrets are held in the cloud provider's managed secret store, not in application code.
Granted, scoped, revocable, logged
Access exists because an administrator granted it to a named invite. Every request is evaluated against the caller's organization on the server. Revocation takes the same path as the grant. Reviewer actions are logged with who, when and why.
Transparency
Who else touches the data, and who to call.
Two classes, named in full on request
Cloud infrastructure: Google Cloud, where the product runs and the data rests. Website analytics: the providers named in the privacy policy, which see this website and never the product. The full subprocessor list travels with the vendor-review packet.
Proven on every release
An automated suite attempts cross-organization reads against every release before it ships. A single failure blocks the deploy. The gate exists so that tenancy is a property of the system, not a promise in a document.
A mailbox a human reads
Security reports and incident questions go to hello@fraudbrain.ai, also published at /.well-known/security.txt. Reports are read by a person and answered.
Request the vendor-review packet.
Architecture, data handling, access model and the full subprocessor list, walked through with your diligence team.
Request an architecture reviewMark your note "vendor review". A human reads every request. Prefer email? hello@fraudbrain.ai