1. Security program
Security is owned by a dedicated team that reports to the CTO and holds a standing veto over releases. Policies are reviewed annually, and every employee completes security training at onboarding and yearly thereafter.
We run an annual third-party penetration test against the production environment and publish the summary to customers under NDA. Findings are tracked to closure with severity-based deadlines.
2. Compliance
Vantafold maintains a SOC 2 Type II report covering security, availability, and confidentiality. The current report is available to customers and prospects under NDA.
We are GDPR-ready and offer a Data Processing Addendum with Standard Contractual Clauses. A completed CAIQ and SIG Lite are refreshed quarterly and sent on request.
3. Encryption
All data is encrypted in transit with TLS 1.3, with TLS 1.2 supported for older clients. HSTS is enforced on all domains. Data at rest is encrypted with AES-256 using keys managed in a hardware security module and rotated annually.
Enterprise customers may supply their own encryption keys for trace storage.
4. Access control
Internal access to production follows least privilege:
- Single sign-on with hardware-backed multi-factor authentication is mandatory for all staff.
- Production access is time-bound, requires a second approver, and is logged.
- No standing access to customer content; break-glass access generates an alert and a review.
- Access is reviewed quarterly and revoked automatically on role change or departure.
5. Infrastructure
The platform runs on hardened container images in isolated network segments across nine regions. Infrastructure is defined as code, and changes go through peer review and automated policy checks before deployment.
Systems are patched on a risk-based schedule: critical vulnerabilities within 72 hours, high within seven days. Images are rebuilt weekly regardless of known findings.
6. Data handling
Customer content is logically isolated per tenant, and prompt prefix caches are never shared across tenants. Zero-retention mode discards prompts and completions entirely, keeping only metadata such as token counts and latency.
Field-level redaction can strip designated fields before anything is written to storage. Backups are encrypted, tested quarterly, and destroyed on the same retention schedule as the primary data.
7. Monitoring and response
We maintain centralised logging with automated alerting for anomalous access, configuration drift, and abuse patterns. Our incident response plan is tested twice a year in tabletop exercises.
Confirmed security incidents affecting customer data are reported to the affected customers within 72 hours, with a written post-incident review to follow.
8. Business continuity
The platform runs active-active across regions, so the loss of a single region does not interrupt service. Recovery objectives are an RTO of one hour and an RPO of five minutes for platform metadata; inference itself fails over without data loss.
Continuity plans are tested annually, and results are shared with Enterprise customers on request.
9. Vendor management
Every subprocessor is reviewed before onboarding and annually thereafter, covering security posture, compliance certifications, and data handling. The current list is published on the subprocessors page, and customers receive 30 days' notice of additions.
10. Vulnerability disclosure
Report vulnerabilities to security@vantafold.ai. We acknowledge within one business day and aim to provide a remediation timeline within five.
We will not pursue legal action against researchers acting in good faith who avoid privacy violations, service degradation, and data destruction, and who give us reasonable time to fix an issue before disclosing it. We credit reporters who want credit.
11. Testing scope
In scope for good-faith research:
api.vantafold.aiand the dashboard, using your own account and test keys.- Our official SDKs and the OpenAPI specification.
- The marketing website.
Out of scope: denial-of-service testing, social engineering of our staff or customers, physical attacks, and findings that require a compromised end-user device. Third-party model providers are governed by their own disclosure programs.
Questions about this document? Write to legal@vantafold.ai, or see the contact page for our registered entities.