How we handle your eval data
Fyntune captures production LLM inputs and outputs for evaluation — which may include user-submitted content, internal documents, and proprietary prompts. We designed our data handling around per-team isolation, 30-day maximum raw data retention, and strict no-training policies. We don't claim certifications we haven't completed. What we describe here is what we've actually built.
Security questions? [email protected]
Security controls designed for production data
We tell you what we've built and how we've built it. We don't claim certifications we haven't completed. Where we say "designed with X controls," that means the control exists — not that a third-party auditor has verified it.
All eval data is encrypted at rest using AES-256. All data in transit uses TLS 1.3. API keys are hashed — Fyntune staff cannot read your API key after issuance.
Each Fyntune team's eval data is logically isolated — no cross-tenant data access is possible by design. Production LLM inputs and outputs captured for evaluation are never accessible to other teams or used in aggregate model training.
Raw LLM inputs and outputs captured during eval runs are retained for a maximum of 30 days, then permanently deleted. Aggregate eval scores and deltas are kept per your plan's retention policy. Enterprise customers can configure custom retention windows.
API keys support scoped permissions (read-only vs read-write) and can be rotated or revoked at any time from the dashboard. Team admins control which users can create API keys. Enterprise tier adds SSO/SAML integration and role-based access control.
LLM inputs and outputs processed through Fyntune's eval pipeline are never used to train, fine-tune, or improve any AI model — Fyntune's or any third party's. This is a firm policy, not a configuration option.
Enterprise customers can configure the geographic region where their eval data is stored and processed. Available regions include US and EU. Data residency agreements are available on request for regulated industries.
Have specific security requirements or need a security review before onboarding your team? Email us at [email protected] — we're happy to walk through our architecture and answer specific questions.