Which part of the Einstein Trust Layer architecture leverages an organization's own data within a large language model (LLM) prompt to confidently…

Dynamic Grounding is the component of the Einstein Trust Layer specifically designed to enhance LLM responses by integrating real-time, contextual data from an organization's own sources, such as Salesforce Data Cloud and CRM records. This process, a form of Retrieval-Augmented Generation (RAG), automatically adds relevant and up-to-date company data to the user's prompt before it is sent to the LLM. This "grounds" the model's response in factual, company-specific information, significantly improving the accuracy and relevance of the generated output.

A. Prompt Defense: This is a security feature that inspects prompts to prevent malicious injections or attempts to extract sensitive data, not for adding contextual data.

B. Data Masking: This is a data privacy feature that automatically redacts or anonymizes personally identifiable information (PII) from prompts before they are processed by the LLM.

1. Salesforce Official Documentation

"Einstein Trust Layer": "To provide LLMs with the context needed to generate relevant outputs

we dynamically ground prompts in your company data. This is a form of Retrieval Augmented Generation (RAG)

which retrieves data from a specified source and adds it to the prompt." (salesforce.com/products/einstein/einstein-trust-layer)

2. Salesforce Developers Documentation

"Einstein 1 Platform": Under the "Einstein Trust Layer" section

it states

"Dynamic Grounding: Grounds the prompt with the right real-time data to generate a more relevant and accurate response." (developer.salesforce.com/einstein-1-platform)

3. Salesforce Help

"Protect Your Data with the Einstein Trust Layer": This document describes the security and privacy features

clarifying the roles of Data Masking ("Mask sensitive data in prompts and responses") and Prompt Defense

distinguishing them from Dynamic Grounding's function of adding context. (help.salesforce.com/s/articleView?id=sf.einsteintrustlayer.htm)