No training by default
Client data is not used to train foundation models unless you explicitly agree in writing. We use vendor APIs with data handling appropriate to business use.
Trust & data
We work with US-based clients on US-friendly hosting. Security and confidentiality are part of delivery—not a slide at the end of the deck.
Geography
Our default engagements are with US-headquartered organizations, US contracts, and infrastructure aligned to US data residency expectations.
We do not offer EU/UK GDPR-grade programs or EU-region hosting as a standard offering today. If your requirements change, we’ll say so clearly during scoping rather than overpromise.
Data handling
Client data is not used to train foundation models unless you explicitly agree in writing. We use vendor APIs with data handling appropriate to business use.
Separate environments per client where practical. Role-based access for Lumera Labs staff; named engineers on production systems.
Retention periods are defined in the SOW—typically 30–90 days after engagement unless you require longer for operations. Deletion on request at end of contract.
High-impact outputs (client-facing replies, external deliverables) require human review before send by default.
Knowledge systems prioritize source citations. We maintain evaluation sets to catch regressions when models or prompts change.
If we confirm unauthorized access affecting your data, we notify you promptly per contract and work with you on containment.
Subprocessors
Depending on your architecture, subprocessors may include cloud hosting and AI API providers—for example Google Cloud, Microsoft Azure, OpenAI, or Anthropic. We maintain a subprocessor list and disclose changes per your DPA.
Exact vendors are chosen during scoping based on your security requirements, existing stack, and pilot scope—not a one-size-fits-all default.
Boundaries