Your findings are yours. Nobody else sees them.
Letting something test your systems only works if the boundary around your data is absolute and the record of what it did is complete. Both are built in, not settings.
Four commitments.
Single tenant by default
Your engagements, credentials and knowledge graph never share a boundary with another customer. There is no shared corpus into which your estate could leak, because the architecture does not have one.
Encrypted at rest and in transit
Credentials sit in a vault rather than in configuration, and the agent's reasoning trail is stored encrypted rather than as plain text, so a database copy is not a transcript of your weaknesses.
Never used to improve a model
Your findings are not pooled with other customers', not sold on as threat intelligence, and not fed back into any shared model. This is written into the agreement, not just the marketing.
You set retention and deletion
Choose the retention window, export what you need in a portable form, and delete an engagement and its graph outright at any time. Deletion removes the evidence too, not just the index.
The agent cannot exceed what you authorised.
Scope is not advisory. Every outbound request is checked against the authorised set before it is issued, and refusals are recorded alongside successes.
GET app.your-domain.com/api/v1/accountsALLOWPOST app.your-domain.com/loginALLOWGET unrelated-third-party.example/BLOCKDELETE app.your-domain.com/api/v1/accounts/1BLOCKTargets are resolved and pinned at provisioning. Requests to anything outside the authorised set are refused rather than warned about.
A target that changes underneath the engagement cannot be used to redirect the agent at infrastructure you never authorised.
Operations that would damage data or disrupt availability are blocked by default, and every block is logged with its reason. Coverage is never traded for uptime.
Every request, decision and refusal is written down in order, so you can prove exactly what was touched and when, to an auditor or to your own board.
Start with one of your assets.
The enquiry form asks only what we need to write a scope: the asset, the surface it sits on, and who authorises testing on it.
GO TO CONTACTNothing you submit is shared outside Cipher, and enquiry details are never used to train a model.