AI data processing and model training

AI features need relevant context to answer or edit. Corvio treats that context and its derived representations as Customer Content, not disposable telemetry.

Updated
On this page
  1. Data used for an AI request
  2. No training on Customer Content
  3. Service analytics are separate

Data used for an AI request

Depending on the feature, Corvio may send the prompt, selected documents, retrieved passages, prior conversation context, and necessary instructions to an enterprise AI inference provider. Embeddings, summaries, memories, and retrieved chunks inherit the protection level of their source content.

No training on Customer Content

Corvio does not use Customer Content to train public or shared foundation models. We require AI providers used for Corvio inference not to train their foundation models on Corvio inputs or outputs.

Provider abuse-monitoring and retention behavior can vary by deployment and model. Corvio reviews the actual configuration before making a zero-retention claim.

Service analytics are separate

Corvio may use aggregated or de-identified operational measurements such as latency, feature usage, token counts, and error codes to operate and improve the Service. These measurements should not contain document text, prompts, or AI answers.