Skip to content

Trust center

This is the consolidated trust summary for Venturi: the single page a prospective customer’s security and procurement reviewer can read to get the answer to every common diligence question, with a pointer to the detailed page behind each answer. Every row here is defensible from the rest of this section.

Rows describe the designed control. The authoritative release-state label for each control (GA, Private preview, Milestone-scoped, or Target-state) lives in Docs authority & product state; identity-enforcement controls in particular are Target-state there.

Trust summary

Trust question Venturi’s answer
Where does our data live? Inside your own VPC, in your own cloud account. Customer transactional and AI-invocation data never leaves your trust boundary. → Security architecture
Can Venturi reach into our network? No. The control plane is outbound-only and region-pinned; it initiates nothing inbound. Full offline mode is available. → Security architecture
Can Venturi change anything in our systems? No. All integrations are read-only, with an explicit IAM Deny on writes; read-only cannot be overridden. → Security architecture
Can Venturi take our AI traffic down? No. The decision-time interceptor is fail-open with a hard 50 ms timeout; on any failure it forwards your request unmodified. Not configurable. Venturi unavailability never blocks your AI traffic; the 99.9% serving SLO bounds attribution availability, not your traffic. → Security architecture
Do you read our prompts/completions? No. Content inspection is disabled by default; Venturi processes invocation metadata only, never bodies, and stores API keys as a truncated prefix. → Data privacy & retention
Do you profile or score individual employees? No. Adoption intelligence is cohort-only (minimum cohort of 5), with sub-cohort suppression; individual-level views are off by default and hard-disabled in the EU. No emotion or behavioral-state inference. → Data privacy & retention
How is our data isolated from other tenants? Per-tenant database, event topics, and a dedicated KMS key per tenant; cross-tenant queries are rejected with 403 TENANT_MISMATCH and logged (record-level enforcement is Target-state, labeled on the tenant-isolation page). → Tenant isolation
How is data encrypted? AES-256 at rest with your KMS key; TLS 1.3 in transit; mTLS (SPIFFE) for internal gRPC; cert-pinned control-plane→data-plane path. → Security architecture
How are secrets managed? A managed secrets store; never on disk or in pod env; database credentials rotate every 90 days; JWT signing keys are externally managed and rotated; your provider keys never reach our control plane in plaintext. → Security architecture
What about audit logs? Append-only audit log, PII pseudonymized, recording every policy decision, model promotion, override, and break-glass access; designed for 5-year Object-Locked retention (compliance-grade lock is Target-state, labeled on the audit-log page). → Data privacy & retention
How long do you keep our data, and can we delete it? Operational retention defaults to 13 months (configurable 30 days–5 years). Erasure is by crypto-shredding your per-subject/per-tenant keys, certified within 30 days, with a deletion certificate. → Data privacy & retention
How do you handle a right-to-erasure request? As your processor, on your instruction: operational/attribution data is crypto-shred-erased within 30 days; audit/override stores are retained pseudonymized under a named lawful basis; anonymized aggregation contributions are non-personal and not individually reversible. → Data-subject rights
Are you SOC 2 certified? SOC 2 Type I readiness/report is a target; Type II is targeted within 12 months. It is not yet held. Scope is Security + Availability + Confidentiality + Processing Integrity (Privacy added on the first EU/regulated deal). → Compliance
What is your GDPR role? We are your processor; you are the controller. We commit Art. 28(3) DPA terms, RoPA, a DPIA template, a lawful-basis mapping, and Art. 28(2) subprocessor change-notification. → Compliance
GDPR / data residency? Data stays in your VPC; the control plane and cross-tenant aggregation are region-pinned from day one (single region + SCCs); EU data stays in-region or under Chapter V safeguards. → Residency & subprocessors
EU AI Act? We document a non-high-risk posture under Art. 6(3)(c)/(d), register under Art. 49(2) regardless, perform no emotion inference (Art. 5(1)(f)), and stay out of profiling via cohort-only adoption intelligence. → Compliance
DPA & subprocessors? Minimal subprocessor surface by architecture (your cloud hosts the data plane). A formal DPA and a published, versioned subprocessor list are produced in the enterprise-readiness workstream. → Residency & subprocessors
Incident response & breach notification? A documented Sev-1/2/3 plan; we notify you without undue delay (target ≤24 h of becoming aware) so you can meet your GDPR Art. 33 72-hour deadline. → Compliance
How is the platform threat-modelled? A STRIDE threat model maps each threat to its control and residual risk, exercised by a red-team program committed to run at least annually. → Security architecture
Can we trust the numbers for chargeback? Operational confidence is an auditable policy ceiling of 0.95 and is never raised; only coper ≥ 0.80 is chargeback-eligible; provisional attributions are labeled and never auto-billed; unknown is reported as unknown, never zero. → Confidence and evidence
What about your supply chain? Dependency and image scanning with critical-finding remediation within 7 days; signed images and an SBOM per release; model artifacts are hash- and signature-verified at load. → Security architecture

Control framework crosswalk

Venturi’s in-architecture controls map to the principal governance frameworks. ISO/IEC 27001 (ISMS) and ISO/IEC 42001 (AIMS) certification are roadmap items; the mapping is provided now to accelerate diligence and show how the control design matures into a managed-system certification.

Venturi control NIST AI RMF NIST SP 800-53 / CSF ISO/IEC 27001 / 42001
Fail-open + per-plane SLOs + error budget MANAGE 2.x; MEASURE 2.x CP-10, SC-5; CSF RC.RP 27001 A.5.29, A.8.14; 42001
Confidence cap / calibration / false-high-confidence gate MEASURE 2.x SA-11; RA-3 42001
Fairness / subgroup-disaggregated evaluation MEASURE 2.11 N/A 42001
Read-only IAM deny / least privilege MANAGE 1.x AC-3, AC-6 27001 A.5.15, A.8.2
Tenant isolation / cross-tenant 403 MAP 5.x SC-2, SC-4, SC-7 27001 A.8.20–A.8.23
Encryption at rest/in transit, KMS lifecycle N/A SC-8, SC-12, SC-13, SC-28 27001 A.8.24
Secrets management / rotation N/A IA-5, SC-12 27001 A.5.17, A.8.24
Audit trail (pseudonymized, object-locked retention) GOVERN 1.x; MANAGE 4.x AU-2, AU-9, AU-11 27001 A.8.15, A.5.28
Retention / crypto-shred erasure / DSAR MAP 4.x SI-12; MP-6 27001 A.5.34; 42001
Threat model / red-team MEASURE 1.x; MANAGE 1.x SA-11, SA-15, RA-3, CA-8 27001 A.8.8, A.8.29
Supply chain / SBOM / signing MAP 4.1 RA-5, SR family; SSDF SP 800-218 27001 A.8.8, A.5.21
Human oversight of model selection & disputes GOVERN 1.x; MANAGE 3.x N/A 42001
EU AI Act self-classification GOVERN 1.x N/A 42001

Human oversight

A single human-oversight requirement governs which decisions require a human: production model selection is a human Venturi-leadership decision recorded with an evidence card (an automated approver is rejected), and customer disputes and overrides are human-adjudicated and recorded to the audit trail with an SLA. Both are recorded in the audit trail.

Reliability commitments

The serving plane carries explicit SLOs, distinct from the fail-open customer-traffic guarantee:

Objective Target
Customer AI-traffic availability Never blocked by Venturi (fail-open by design)
Serving plane (query/dashboard API) availability 99.9%
Recovery point / time objective RPO ≤ 15 min / RTO ≤ 1 h
Index freshness P99 ≤ 90 s
Reconciliation latency ≤ 24 h
Breach notification to you Target ≤ 24 h of becoming aware

A Business Continuity Plan extends disaster recovery to key-person succession, cloud-provider or region outages, and critical-subprocessor failure, and is exercised at least annually.

Diligence artifacts & vulnerability disclosure

  • Diligence packet. A security and data-processing diligence packet (covering the trust-boundary model, the control matrices, the threat model, and the lawful-basis mapping) is available to prospective customers under NDA. Contact your Venturi representative or reach the security team to request it.
  • Vulnerability disclosure. The security program operates a coordinated Vulnerability Disclosure Program with a security.txt (RFC 9116) monitored intake, safe-harbor for good-faith research, and coordinated-disclosure terms. Report a suspected vulnerability privately to security@venturi.systems.
  • Status page. A public status page reports component health and active or scheduled-maintenance incidents.