Einstein Trust Layer

Question 1

Zero-Data Retention Policy

Answer 1

Your data isn’t retained by third-party LLMs. Salesforce partners with Open AI and Azure Open AI to enforce the zero-data retention policy.

Question 2

Dynamic Grounding with Secure Data Retrieval

Answer 2

Secure data retrieval of Salesforce data for grounding the prompt based on the permissions of the user executing the prompt.

https://help.salesforce.com/s/articleView?id=sf.generative_ai_trust_arch.htm&language=en_US&type=5#:~:text=Secure%20Data%20Retrieval%20and%20Grounding&language=en_US

Question 3

Prompt Defense

Answer 3

System policies help limit hallucinations and decrease the likelihood of unintended or harmful outputs by the LLM.

https://help.salesforce.com/s/articleView?id=sf.generative_ai_trust_arch.htm&language=en_US&type=5#:~:text=the%20generated%20response.-,Prompt%20Defense,-To%20help%20decrease&language=en_US

Question 4

Data Cloud Feedback as Part of Audit Trail

Answer 4

Collected with the Feedback API:

1. Thumbs up and thumbs down reactions and reason text for each response
2. Whether a prompt response is accepted, regenerated, modified, declined or ignored
3. Modified final response used
4. Explicit Feedback Used
   (No user data is included in feedback)

Question 5

Data Masking

Answer 5

Sensitive data is detected and masked before sending the prompt to the LLM.

Data masking supports multiple regions and languages.
You can select what must and must not be masked.

https://help.salesforce.com/s/articleView?id=sf.generative_ai_mask_data.htm&language=en_US&type=5

Question 6

Audit Trail Feature

Answer 6

The Einstein Trust Layer provides the ability to audit, track, and view masked data within prompts.

https://help.salesforce.com/s/articleView?id=sf.generative_ai_feedback.htm&language=en_US&type=5

Question 7

Toxicity Scoring

Answer 7

Einstein Trust Layer scores content on toxicity.

Toxicity scores are logged and stored in Data Cloud as part of the audit trail.

https://help.salesforce.com/s/articleView?id=sf.generative_ai_audit_toxicity.htm&language=en_US&type=5

Question 8

Toxicity Score - Safety Category

Answer 8

Safety category ranges from 0 through 1 with 1 being the safest.

Question 9

Set Up Einstein Generative AI

Answer 9

Step 1: Verify Data Cloud in Your Org
Step 2: Turn On Einstein Generative AI
Step 3: Set Up Einstein Trust
Step 4: Turn On Einstein Data Collection and Storage

https://help.salesforce.com/s/articleView?id=sf.generative_ai_enable.htm&type=5

Question 10

Einstein Generative AI and Feedback Data Dashboard

Answer 10

Monitor the use of generative AI within your Salesforce org.

Visually Analyze:
- user feedback,
- token consumption,
- request counts,
- and user engagement weekly

https://help.salesforce.com/s/articleView?id=sf.generative_ai_feedback_analytics.htm&type=5

Question 11

Einstein Trust Layer Dashboard

Answer 11

Monitor Einstein Trust Layer metrics such as data masking and toxicity.

• Data masking trends over the past 7 days, 30 days, and for all time.
• Data masking trends by feature over the past 7 days, 30 days, and for all time.
• Trends in toxic responses categorized over the past 7 days, 30 days, and for all time.
• Trends in toxic responses by feature over the past 7 days, 30 days, and for all time.

https://help.salesforce.com/s/articleView?id=sf.generative_ai_feedback_analytics.htm&type=5

Question 12

Toxicity Detection in Responses Report

Answer 12

A detailed view of the toxic responses detected by the Einstein Trust Layer.

Use it to see the response text and the number of toxicity detection by feature.

https://help.salesforce.com/s/articleView?id=sf.generative_ai_feedback_analytics.htm&type=5

Question 13

Masked Prompt and Response Report

Answer 13

A detailed view of the prompt with the masked data, the full prompt without the masking, and the response with and without the masked data. Use it to see what data was masked before sending it to the large language model (LLM).

https://help.salesforce.com/s/articleView?id=sf.generative_ai_feedback_analytics.htm&type=5