Question

An administrator is enabling a generative AI feature that sends prompts containing customer information to a large language model. Compliance is concerned that personally identifiable information (PII) such as phone numbers and Social Security numbers could be exposed to the third-party model provider. Which capability of the Einstein Trust Layer directly addresses this concern?

Accepted Answer

Data masking, which detects and replaces sensitive data with placeholder tokens before the prompt is sent to the LLM

Answer

Zero data retention, which deletes the LLM response after it is displayed to the user

Answer

Prompt grounding, which enriches the prompt with related Salesforce record data

Answer

Toxicity scoring, which evaluates generated responses for harmful language