What the models see, and what they keep.

The AI-specific commitments that a general privacy policy does not cover: which providers process conversation content, what is actually sent to them, what they retain and for how long, whether anything is used to train a model, and what a caller is told about talking to an agent.

The commitments

• Customer content is not used to train any model, by us or by our providers.
• Providers are engaged under enterprise terms with zero-retention where offered.
• Every provider in the chain is named on the sub-processor list.
• An agent will never deny being AI when asked directly.

What is actually sent to a model

The conversation turn, the relevant retrieved passages from your knowledge base, and the system prompt. Not your whole knowledge base, not your customer database, and not prior unrelated conversations. Retrieval scope is a configuration decision with a privacy consequence, and it is documented per build.

Redaction in regulated builds

Under healthcare builds, protected health information is redacted before transcripts reach any logging or analytics system, and the retrieval index is scoped so an agent cannot surface one patient's details to another caller. Those controls are part of the build, not an add-on setting.