No. Customer data is never used to train Anthropic, OpenAI, or Google models — and that is a contractual obligation under enterprise Zero Data Retention (ZDR) agreements, not a setting that can be flipped. Your claim data flows through ELIAS AI for processing in the moment, but the providers do not retain it, do not log it, and do not use it for any training purpose. Evolution Global also does not train its own models on customer data; we use the providers' commercial models and govern them through the Certified Integrity Standard published by the Ark Alliance Foundation. This commitment is published on the Trust Center and verifiable in the Master Service Agreement.

ZDR agreements with Anthropic (Claude), OpenAI, and Google are signed at the enterprise level. Under those agreements, the provider:

ZDR is a stricter posture than standard "we don't train on your data" language. Many providers' standard terms exclude API customer data from training but still allow short-term retention for abuse monitoring or service improvement. ZDR removes those exceptions. It's the contractual highest-water-mark commitment available from each provider.

Two reasons.

Settings can be changed. A toggle marked "do not train on my data" is dependent on the provider's interpretation, the customer's awareness, and the absence of bugs that could flip the toggle without the customer noticing. Contractual ZDR isn't a toggle — it's a binding agreement at the enterprise level that covers every API call from FileTrac.

Brand commitment. The Ark Alliance Foundation's seven governing principles include "AI Must Serve Justice, Not Convenience" — and that means AI in P&C insurance can't be built on customer data without explicit consent and structural protection. ZDR is the structural protection.

Evolution Global does not train AI models on customer data. We use commercial models from Anthropic, OpenAI, and Google and govern them through the Certified Integrity Standard. There's no "Evolution Global proprietary model" being built from customer claim files.

The reason: building a proprietary AI model on aggregated claim data would create exactly the problem the Foundation was set up to solve — the Integrity Deficit. If Evolution Global built a model on the existing claim data flowing through FileTrac, that model would inherit two decades of suppressed claims data (the "Scrub") that the industry has been training on. The result wouldn't be better; it would just be more of the same drift, automated.

The Golden Record — the clean, CIS-validated benchmark database governed by the Foundation — is the alternative. It's built one CIS-validated claim at a time, scrubbed of nothing, owned by the Foundation, accessible only with Foundation approval. Models trained on the Golden Record are trained on validated reality, not on the historical drift.

The commitment is published in three places:

Penetration test reports and the SOC 2 Type II report (when issued — audit window H2 2026) are available behind the Drata Trust Center under NDA.

ELIAS Concierge is the public-facing chat experience for prospects. Conversations on ELIAS Concierge are governed by the same ZDR agreements; the providers don't retain, log, or train on what you ask the Concierge. The Concierge draws from published source material (the website, the FAQ articles, the Trust Center, the Privacy Policy) — not from any internal claim data.

Evolution Global retains a conversation history for the website-side Q&A experience continuity, but that history is not used to train AI models — yours or anyone else's.