The Einstein Trust Layer is a secure AI architecture integrated into the Salesforce platform. It includes a set of features and guardrails designed to protect data privacy and security, improve AI result accuracy, and promote responsible AI use. This article presents an Overview of Salesforce Einstein Trust Layer.
Key features include:
- Zero-Data Retention Policy
Salesforce’s zero-data retention policy ensures that no data shared with third-party language models is stored or used beyond its immediate purpose. Salesforce enforces this policy with these principles:
- No Data Used for Training: Data sent to third-party LLMs is not used for model training or product enhancements.
- No Data Retention: Once processed, data is not retained by third-party LLMs.
- No Human Review: Data shared with third-party providers is not accessible to their human personnel.
2. Dynamic Grounding with Secure Data Retrieval
The Einstein Trust Layer securely retrieves relevant Salesforce records to provide context for AI prompts, keeping all user permissions intact:
- Secure Contextual Data: Salesforce data is retrieved based on the user’s permissions to personalize responses.
- Role-Based Controls: Salesforce’s standard permissions, including role-based access and field-level security, are preserved when data is used to ground AI responses.
3. Prompt Defense
To ensure accurate, safe responses, system policies within the Einstein Trust Layer guide generative AI behavior:
- Limiting Hallucinations: Policies reduce the chance of unintended or erroneous outputs.
- Context-Dependent Policies: Customizable system policies adapt to the requirements of each AI feature or use case.
4. Data Masking
Sensitive information is automatically masked before it is shared with the AI model, enhancing privacy across global regions:
- Comprehensive Data Masking: Detects and masks sensitive data in prompts, compatible with various regions and languages.
- Flexible Masking Controls: Organizations choose which data fields to mask, providing flexibility in managing sensitive data.
5. Toxicity Scoring
To maintain a respectful, safe environment, the Trust Layer evaluates content toxicity:
- Toxicity Assessment: Scores for potentially toxic content are calculated and stored in Data Cloud for oversight.
- Audit Trail: Toxicity scores are logged as part of Salesforce’s comprehensive audit trail.
6. Audit
For comprehensive oversight, the Einstein Trust Layer tracks all AI interactions:
- Prompt and Response Logging: Both prompts and responses, along with trust metrics, are logged in Data Cloud.
- Pre-Built Reports: Includes reporting and dashboard tools to analyze trends, helping refine prompt templates and AI interactions.
This suite of features ensures a trusted, secure experience with AI, enhancing privacy, transparency, and control.
In this blog, we learned an overview of Einstein Trust Layer. To learn more click Einstein Audit and Feedback Data.
Take Five Consulting is a technology company, based in Virginia U.S., that specializes in the Mortgage Banking vertical especially LOS implementation and application development. Take Five Consulting creates and implement mortgage technology and software specifically for Mortgage Industry.
