Security, reliability, transparency.
ProvableCORE® is engineered for institutional buyers who cannot afford to guess about their infrastructure. This page documents every layer of our security, availability, data handling, and operational posture. What you see here is what auditors see.
Current status
Design philosophy
Built on primitives that already earned institutional trust.
ProvableCORE® was not designed to invent new cryptography. It was designed to assemble components that have been battle-tested for decades — Merkle trees (RFC 9162, in production since 2013), RSA-PSS (FIPS 186-4), TLS 1.3, GCP KMS Hardware Security Modules, WORM storage — into a governance kernel that banks and regulators can verify themselves, without trusting us.
Every component on this page is either a published standard, a certified service from a Tier-1 cloud provider, or a verifiable open-source library. We do not ask you to trust us. We ask you to verify the record.
Infrastructure and data residency
| Layer | Provider | Region | Certification |
|---|---|---|---|
| Compute | Google Cloud Run | europe-west1 (Belgium) | ISO 27001, SOC 2, ISO 27017, ISO 27018 |
| Key management | GCP KMS | europe-west1 | FIPS 140-2 Level 3 HSM |
| Object storage | GCS Bucket Lock (WORM) | europe-west1 | Object retention policy enforced at platform level |
| Event journal | Cloud Firestore | eur3 (multi-region EU) | Append-only via IAM policy + CMEK |
| Analytics | BigQuery | EU | Column-level policy tags, CMEK |
| Identity / Secrets | GCP IAM + Secret Manager | Global (EU replicated) | Workload Identity Federation (keyless CI/CD) |
All customer data — credit events, decision records, signatures — is stored and processed exclusively in EU regions. No transfer to US or third countries occurs without explicit contractual provision (DPA + SCC).
Cryptographic architecture
SHA-256
FIPS 180-4 standard. Used for event leaf hashes, Merkle tree construction, and state snapshot hashes. No legacy SHA-1 anywhere in the system.
RSA-PSS-SHA256 (4096-bit)
Per RFC 8017 PKCS #1 v2.2. Private key generated inside GCP KMS HSM; never exportable, never accessible by our staff. Signing performed via KMS API.
Merkle Tree (RFC 9162)
Certificate Transparency v2 compatible. SHA-256 node hashing. Batch compaction with signed tree head. Any third party can verify an inclusion proof in O(log N) without access to our infrastructure.
SD-JWT (RFC 9901)
Evidence packs can be issued with per-audience redaction profiles. Regulators receive full detail; counterparties receive only what they need. Cryptographic binding preserved.
TLS 1.3
EV SSL certificate via Sectigo. Perfect Forward Secrecy. HSTS enforced. Mozilla Observatory grade A+ target.
GCS Bucket Lock
Object retention policy enforced at Google infrastructure level. Once locked, no IAM principal — including Google Support — can delete or modify an object before retention period expires.
Access control and key management
- Workload Identity Federation: CI/CD pipelines authenticate via OIDC tokens. No long-lived service account keys in GitHub secrets.
- Principle of least privilege: IAM roles scoped to minimum required; no Owner or Editor roles for service accounts.
- Secret rotation: Quarterly rotation schedule for all API keys, SMTP credentials, and third-party integrations.
- Hardware-backed MFA: All human operators use FIDO2 security keys (YubiKey or equivalent) for GCP Console access.
- No root SSH: VM access via IAP (Identity-Aware Proxy) + SSH certificate. No static public IP, no port 22 exposed.
- Private customer keys: Per-tenant CMEK (Customer-Managed Encryption Keys) available on Enterprise tier.
Availability and performance
| Metric | Trial tier | Professional tier | Enterprise tier |
|---|---|---|---|
| Uptime SLA | Best-effort | 99.5% monthly | 99.9% monthly |
| Verify endpoint p50 latency | ≤ 200ms | ≤ 100ms | ≤ 50ms |
| Receipt generation p99 | ≤ 2s | ≤ 500ms | ≤ 300ms |
| Support response time | Best-effort | Business hours (EU) | 24/7 with on-call rotation |
| Incident communication | Status page | Status page + email | Status page + email + dedicated Slack |
Compliance and certifications
| Framework | Status | Scope |
|---|---|---|
| EU AI Act (Annex III high-risk) | Ready | 10 QMS documents available: Technical Documentation, Risk Management, Human Oversight, Conformity Assessment |
| EBA GL/2020/06 (Loan Origination) | Compliant | Milestone-based credit assessment, full model explainability, human oversight layer |
| DORA (2022/2554) | Compliant | Immutable audit journal, cryptographic integrity verification, ICT incident audit capability |
| GDPR Article 35 (DPIA) | Complete | Full DPIA completed April 2026; available on request under NDA |
| ISO/IEC 42001:2023 (AI Management) | In progress | Gap analysis complete; external audit targeted Q3 2026 |
| SOC 2 Type II | In progress | Type I audit targeted Q4 2026; Type II 2027 |
| Basel II / SR 11-7 (Model Risk) | Documentation ready | Decision Proof Objects + Replay satisfy model validation requirements |
Intellectual property
- AGEPI Patent Nr. 7671 — ProvableCORE® core system, filed 14 April 2026
- AGEPI Patent Nr. 7672 — SentientROUTER™ governance control plane, filed 17 April 2026
- USPTO provisional applications — in preparation, filing target 10 June 2026
- EPO / PCT phase — planned following USPTO filing
- ProvableCORE® registered trademark (filed 15 April 2026)
Incident response
Our incident response follows the AIR v1.0 protocol (Autonomous Incident Response): multi-LLM consensus validation, ProvableCORE-signed incident records, human approval gate for action triggers. This means every incident response action is itself auditable and reproducible.
- Critical severity: initial response ≤ 15 minutes, communication ≤ 30 minutes
- High severity: initial response ≤ 1 hour, communication ≤ 2 hours
- Post-incident reports within 5 business days of resolution
- Root cause analysis publicly documented for customer-affecting incidents
- No security incidents to date during controlled evaluation (v0.6.0 since April 2026)
Data processing
- Customer data is never used for model training or any purpose other than delivering the contracted service.
- Sub-processors: Google Cloud Platform (infrastructure), Sectigo (SSL), transactional email provider. Full list available in DPA.
- No third country transfers without explicit Standard Contractual Clauses.
- Right to audit: Enterprise tier customers may audit our controls once per 12 months with 30 days notice.
- Data export: On contract termination, all customer data exported in structured format within 30 days; proof of deletion provided.
Responsible disclosure
Found a security issue? We want to hear about it. Email security@tfholding.md with details. We commit to:
- Acknowledge receipt within 48 hours
- Provide a preliminary assessment within 5 business days
- Keep you informed of progress toward resolution
- Credit you publicly (with your consent) once the issue is resolved
- Not pursue legal action for good-faith security research conducted within our disclosure policy
Open verification
You do not need to trust this Trust Center. You can verify any ProvableCORE® receipt independently, offline, without contacting us. The verification endpoint is public and permanently available:
POST https://provablecore.eu/governance/verify
Open-source verifier CLI planned for Q3 2026 — you will be able to verify any receipt locally with no network call and no dependency on our infrastructure.
Want to review our full security package?
Enterprise customers and prospects can request our Security Architecture Document, DPIA summary, and sample audit reports under NDA.
Request security package View compliance matrix