Control catalogue

Every shipped capability has a stable code, so a reviewer can refer to it, and a mapping to the frameworks buyers score against. Mapped means a control supports a requirement. It never means certified.

23Shipped controls
5Sections
3Frameworks mapped
A

Detection

6 controls
CodeControlOWASP LLM Top 10NIST AI RMFISO 27001 Annex A
SC-01Sensitive-data detection: 31 classes via rules, in-house entity models, a vision model and a verifier, plus a semantic layer for nine categories of stated disclosureLLM02MeasureA.8.12
SC-02National identifier validators: TCKN checksum, VKN, IBAN mod-97, LuhnRejects look-alike numbers.LLM02MeasureA.8.12
SC-03Telecom identifiers: IMSI, ICCID, IMEI, MSISDN, coordinatesLLM02MeasureA.8.12
SC-04Verifier gate: a second model re-checks every candidate before actionLLM02MeasureA.8.12
SC-05Tenant rules: custom classes per tenant or global, activated at runtimeLLM02GovernA.5.12
SC-06Document and image protection in place: PDF, DOCX, XLSX, PPTX, PNG, JPG and scanned pagesSignatures, identity documents and faces are blocked, never sent. Unreadable input is refused.LLM02ManageA.8.12
B

Protection and restoration

6 controls
CodeControlOWASP LLM Top 10NIST AI RMFISO 27001 Annex A
SC-07Context tokenisation: opaque per-case tokens, one token per entityLLM02ManageA.8.11
SC-08Format-preserving encryption mode, opt-in, key held by the customerLLM02ManageA.8.24
SC-09Restoration inside the perimeter, scoped to the caseLLM05ManageA.8.11
SC-10Fail-closed: protection failure or unreadable input stops the requestLLM02ManageA.8.12
SC-11Gateway guardrail: streaming protect and restore for gateway-fronted trafficLLM02LLM05ManageA.8.12
SC-12Endpoint deployment with human review before sendLLM02ManageA.8.12
C

Tenancy, identity, operations

4 controls
CodeControlOWASP LLM Top 10NIST AI RMFISO 27001 Annex A
SC-13Tenant-scoped API keys and organisation isolationLLM10GovernA.5.15A.8.3
SC-14Rate limiting and distributed ledgerLLM10ManageA.8.6
SC-15Administrator console, local accounts, API-key lifecycle—GovernA.5.16
SC-16Reviewer feedback loop: corrections captured, never silently retrainedLLM02MeasureA.8.16
D

Evidence and audit

3 controls
CodeControlOWASP LLM Top 10NIST AI RMFISO 27001 Annex A
SC-17Security event logging: ECS JSON with stable codes to your SIEM, no values reprinted—GovernA.8.15
SC-18Durable audit store of detection runs (PostgreSQL)—GovernA.8.15
SC-19Service metrics (OpenMetrics)—MeasureA.8.16
E

Deployment and supply chain

4 controls
CodeControlOWASP LLM Top 10NIST AI RMFISO 27001 Annex A
SC-22In-perimeter delivery: private OCI Helm chart, six images, OpenShift 4.x, mirrored-registry installLLM03GovernA.5.23A.8.9
SC-23Signed release manifest and CycloneDX SBOM per releaseLLM03GovernA.5.21
SC-24No Salus-side data, keys or telemetryLLM02GovernA.5.23
SC-25CPU-only inference, no GPU dependency—ManageA.8.6

Administration and operations

Administration console

Web console in English and Turkish. Detection classes on and off, custom rules by keyword or pattern, API keys per calling system, review screen with CSV export.

Your directory, your roles

Administrators authenticate against Active Directory or LDAP; rights follow group membership. Three roles: administrator, rules editor, reviewer. No local privileged accounts, no shared logins. Single sign-on on request.

Policy per calling system

Each caller gets its own key and its own policy: tokenise, suppress, allow or block, per class. Deterministic, versioned, and rolled back with the release.

Tamper-evident audit trail

Write-only: every detection, restoration, configuration change, key rotation and login attempt, with hashes instead of content. Exported as JSON lines with a CEF mapping to your SIEM.

Monitoring you already have

OpenMetrics on every service: request rate, p50 and p95 latency, error rate, detector availability. A Grafana dashboard and a default alert set ship with the release.

Sized for a CPU node

Reference production node: 16 cores, 32 GB RAM, no GPU. Stateless services with replicas for availability; a disaster-recovery instance is covered by the licence.

Models and rules are versioned artefacts pinned to each release. A release is either active or rolled back; nothing retrains silently.

Every control listed is shipped. KVKK and EU AI Act mappings are provided with the datasheet.

Deliberately out of scope: prompt-injection classification, agent and tool control, retrieval access control, brand and toxicity filtering, red teaming of models. Salus is the data-transformation layer under those tools, not a replacement for them.