Security

Security at QOREN

Security is not a page on this site — it is a set of controls the platform enforces on every request. Here is what each one does, and how you can check it.

Controls that are enforced, not promised

Every control below is implemented in the request path, not described in a policy document.

Zero-trust RBAC and per-tenant isolation
Role-based access control and tenant isolation are applied on every request. A tenant can never read another tenant’s records, and the check happens before the data is fetched.
AES-256-GCM encryption
Data is encrypted in transit and at rest. Credentials are held server-side in an encrypted vault and are never returned to the browser.
Append-only audit trail
Every read, write and tool call is recorded with the actor, the tenant and the result. The log is append-only, so an entry cannot be edited after the fact.
Human approval before any write
Reads run automatically. Anything that changes a record is queued for a named human, and nothing bypasses that gate.
Prompt-injection and output screening
Untrusted input is inspected before it reaches a model, and generated output is screened before it is returned. Injection attempts are rejected and logged.
Least privilege by default
Connections start read-only. Write access is a separate, explicit grant — so a compromised integration still cannot change your systems.

The security posture, in numbers

0
Unapproved writes — every write needs a human
45
Connectors, each with its own scoped credential
14
Governed agents, all audited
AES-256
Encryption in transit and at rest

Questions a security reviewer asks first

Where are credentials stored?

Server-side, in an encrypted vault, scoped per tenant and per connector. They are never sent to the browser and never appear in an API response.

Can one tenant see another tenant’s data?

No. Tenant isolation is enforced on every request, before the query runs. A request without the right tenant context is rejected rather than filtered.

Can an agent change my systems without approval?

No. A write is prepared, queued and shown to a named person with the risk level attached. Nothing is executed until it is approved, and the approval is audited.

What happens to the data you send to a model?

Prompts are screened for injection attempts before they are sent, and responses are screened before they are returned. The public copilot on this website uses public product facts only and never sees visitor or tenant data.

How do I verify the audit trail is complete?

Every action writes one entry with the actor, the tenant, the timestamp and the outcome. The log is append-only, and entries can be exported for an external review.

Get the security pack

The architecture note, the control list and the audit format — sent as one document you can hand to your reviewer.