Einstein Trust Layer in Salesforce Agentforce

Salesforce has launched Agentforce, which allows users to ask questions to agents, who then perform tasks and retrieve information.

Information must be secure between retrieval from the server and display on the agent UI. Therefore, Einstein’s trust layer plays a crucial role here.

In this tutorial, we will learn about the Einstein trust layer in Salesforce Agentforce. In that, I will explain how it protects data and how to enable the data cloud and Einstein Trust Layer in Salesforce.

Table of Contents

Einstein Trust Layer in Salesforce Agentforce

The Einstein Trust Layer is a secure AI architecture integrated into Salesforce. It plays an important role in protecting company and customer data while working with Einstein generative AI.

In the current era of AI, protecting sensitive information is crucial; this sensitive information could include customer data with private details.

The Einstein trust layer encrypts this data by using data masking to prevent sensitive information from reaching LLM. So, it takes care of the data masking to mask the sensitive data.

Apart from masking the sensitive data, Salesforce also has a zero data retention policy with these LLM providers.

This policy requires LLM to forget all the Salesforce data after an AI response has been generated. That means they cannot retain the Salesforce data and cannot use it to train their LLM models further.

What is the LLM Model?

In AI, a Large Language Model (LLM) is a type of AI program trained on large amounts of text data to understand and generate human-like text, used for tasks like natural language processing, translation, and content creation.

Many modern LLMs can remember previous parts of a conversation or document, improving relevance in long-form interactions. However, Salesforce doesn’t allow us to remember after generating an AI response.

How does Einstein Trust Layer Protect Data in Salesforce?

The diagram below helps us understand how the Einstein trust layer protects the data while working with Gen AI.

Einstein Trust Layer in Salesforce Agentforce

1. Prompt & Secure Data Retrieval:

When Salesforce needs to invoke these LLM providers while working with Einstein Copil ot or a prompt template, it may have to send the prompt to these LLM providers so that the Einstein trust layer can first ensure that the data is securely retrieved from your CRM.

2. Dynamic Grounding:

Then, it performs dynamic grounding, which means Salesforce data needs to be added to the prompt so that we can get a personalised response from the LLM model.

3. Data Masking:

Before this final prompt is sent to the LLM, the grounded prompt template is analysed for personally identifiable information, which is then masked.

For example, an email address that will be masked so that the actual email address is not sent to the LLM. When the response is received, the data is masked and removed through data demasking, allowing the end user to view only the necessary information.

4. Secure Gateway & LLM Providers:

After data masking, the prompt moves through the LLM secure gateway and goes to external LLM providers like OpenAI and ChatGPT for processing. Salesforce has a zero data retention policy with these providers, which means that after the LLM has processed the prompt and returned with the response, the LLM provider won’t store any data referenced.

This ensures that Salesforce data is always handled securely and never retained externally.

5. Response Generation & Data Demasking:

Once the LLM generates the response and sends it back to Salesforce, the Einstein trust layer first removes the data masking and then performs toxicity detection.

6. Toxicity Detection:

The toxicity detection feature reviews the response and determines a toxicity score. This helps to identify whether a response includes harmful content that may be unsafe for external use.

The toxicity score ranges from 0 to 1, where zero means the response is not toxic, and one means it is the most toxic.

Similarly, a safe score is also assigned from 0 to 1, where zero is not safe and one is safer. This is the opposite of the toxicity score.

7. Audit Trail:

The trust layer also keeps an audit trail, so we can always track AI use within the organisation. Finally, the final response is shown to the end user.

Therefore, we can say that the Einstein trust layer is a closed-loop system that protects company and customer data throughout the generative AI life cycle, also helping to improve the accuracy and safety of AI within Salesforce.

Setup Einstein Trust Layer in Salesforce

As the Einstein trust layer masks the sensitive data, the Einstein platform determines which sensitive data should be masked. But recently, Salesforce has allowed organisations to configure sensitive data.

In the steps below, I will explain how to set up the Einstein trust layer to configure the data.

1. Enable Data Cloud in Salesforce

If you didn’t even enable the Einstein Setup, we need to enable it first, then the Einstein Copilot for Salesforce, which is Agentforce.

Now, to set up the Einstein trust layer, we need to have a data cloud-enabled; only then can we use the Einstein trust layer.

To enable the data cloud, you need to have a data cloud admin permission set. For that, navigate to the user settings.

Then go to the Advanced User Details and click the Permission Set Assignments. After that, click on the Edit Assignment button.

Admin Permission Set Assignments in Salesforce

Then, search for the Data Cloud Admin permission set and click the Add button to assign that permission set to yourself. Then click the Save button and refresh the page to see the Data Cloud step option.

Data Cloud Admin Permission in Salesforce

Now, as you click the Gare icon, you will see the Data Cloud Setup option available. Click on it.

Then, scroll down to Setup Data Cloud. Then, click on the Get Started button. It will start the setup of a data cloud instance in this org, which is called the home instance.

The data cloud will be available within the same Salesforce org where all of your Salesforce data is available.

This process will take a little time to complete the setup. It will take 25 to 30 minutes.

After enabling the data cloud in this org, you can check by clicking on App Launcher and searching for the Data Cloud app. You will see that the app is enabled.

2. Setup Einstein Trust Layer in Salesforce

After enabling the data cloud setup, we are ready to set up the Einstein trust layer for data masking.

For that, in the Quick Find, search for the Einstein Trust Layer and click on it. After that, you will see the Large Language Model Data Masking option.

We need to enable it. To do so, simply click the toggle button. This will allow us to configure the data masking fields.

Setup Einstein Trust Layer in Salesforce

Now, to configure the sensitive data, scroll down, and you will see the Sensitive Data option.

Here, you can see the default field that the Einstein trust layer uses to mask the data; it will consider the field specific to the organisation.

From this screen, you should be able to add the field that the Einstein trust layer should consider for masking when it is about to send the data to the external LLm providers.

In this way, we can set up the Einstein trust layer in Salesforce Agentforce.

Conclusion

I hope you have got an idea about the Einstein trust layer in Salesforce Agentforce. In that, I have explained how the Einstein trust layer protects the data, the features of every layer, and how to enable the data cloud and Einstein trust layer in Salesforce to configure the sensitive information to mask the data.

Einstein Trust Layer in Salesforce Agentforce