Trust, privacy, and compliance

How the Privacy Vault protects sensitive data

In brief: The Privacy Vault filters and tokenizes sensitive data before it reaches a third-party generative model. The model receives filtered text and tokens, not raw personal identifiers.

Sensitive data can appear in customer messages and other content that an agent uses to answer a request. The Privacy Vault protects that information as it moves through Unless, while allowing identifiers to be restored inside the controlled Unless environment when needed.

How the Privacy Vault works

The Vault sits at the edge of Unless, where data enters the platform. It detects and filters personal data at ingestion and again at runtime, before a call to a third-party generative model. This means names, email addresses, phone numbers, and other identifiers are removed or masked by default.

When a response needs personal context, the Vault can replace sensitive values with tokens. The agent works with those tokens, while the Vault alone holds the mapping to the original identifiers. An authorized component can de-tokenize a value when needed, such as to address a customer by name.

What the model receives

The model receives scrubbed input, not recoverable personal identifiers. It generates an answer from that filtered text and any tokens, and the Vault re-attaches identifiers only when the response requires them.

The PII-detection step itself may process personal data in the clear. It is identified as a sub-processor in the DPA; the model that phrases the final answer does not see the raw values.

What you can configure

Privacy controls let you adjust how personal data is handled without changing the Vault’s role in protecting it. Depending on your setup, you can:

• Configure PII filtering and add fields to a whitelist. Whitelisted fields are shown to the agent in plain text, so add only fields the model needs.
• Set whether tracking and personalization require explicit consent, including whether cookieless experiences can appear before consent.
• Process data access, correction, and erasure requests through built-in workflows.

Where data is handled

Personal data on the platform and the Vault stay in the EU. Platform sub-processors are EU entities of AWS, Microsoft Azure, and Google Cloud, and their platform processing takes place inside the EU. Unless’s business tools for its own administration may be outside the EU, but they do not handle end-user data.

What happens to customer data

Unless does not use customer content, end-user conversations, or other customer data to train models. Living Knowledge improves through approved content and Team Feedback instead.

The Privacy Vault protects sensitive identifiers as they pass through the platform, while configuration and request workflows support your privacy processes. Together, these controls help keep raw personal data away from the generative model.