Secure Data Retrieval allows users to access data securely, grounding generative AI prompts in the context of your business while maintaining strict permissions and data access controls.

Example:

Imagine you're in the marketing department and want to use AI to generate personalized email campaigns. You define the specific data points needed for the AI, like customer purchase history and demographics. Salesforce verifies your access rights, ensuring you retrieve only the data you’re authorized to see, thereby protecting confidential information from unauthorized users.

Data Masking safeguards sensitive data by masking personal identifiable information (PII) and payment card industry (PCI) information before sending AI prompts to third-party large language models (LLMs).

Example:

Your sales team wants to use a third-party LLM to generate personalized sales pitches. Before sending data to the LLM, Salesforce masks sensitive fields like customer names and credit card details. This ensures the LLM can analyze customer profiles and generate content without seeing actual sensitive information.

Prompt Defense helps limit hallucinations and decreases the likelihood of unintended or harmful outputs by the LLM through system policies.

Example:

You’re using a Salesforce AI tool to generate product descriptions. A prompt might lead the AI to fabricate information, such as claiming a water bottle can "purify ocean water with a thought." Prompt Defense analyzes the prompt against predefined policies and alerts you with a warning, suggesting revisions to avoid such unrealistic claims.

Zero Retention ensures that your data isn't retained by third-party LLMs. Salesforce partners with OpenAI and Azure OpenAI to enforce this policy.

Example:

You’re writing a confidential email using a platform that utilizes an LLM for spell check. The content of your email isn’t saved on the servers of third-party LLM partners and isn’t used to train or improve their models. No employee at third-party providers sees the content, ensuring complete privacy.

Toxicity Detection through the Einstein Trust Layer scores content on toxicity, ensuring generated content is appropriate and avoids offensive language.

Example:

You're managing a social media marketing campaign and want to ensure the content generated by an LLM is appropriate. The Einstein Trust Layer analyzes the generated content for harmful or offensive language, assigns a toxicity score, and flags potentially offensive content for review before posting.

Audit Trail logs and stores prompts, responses, and trust signals in Data Cloud, providing pre-built reports and dashboards for analysis.

Example:

Example: You're managing a lead generation campaign powered by an LLM. The LLM interacts with potential customers through chatbots. The Audit Trail captures all interactions, allowing you to analyze which prompts and responses lead to the most conversions. You can then adjust your strategies based on this data to improve lead qualification.

By integrating these AI capabilities into Salesforce, businesses can enhance their operations while ensuring data security and integrity. Trust in AI within Salesforce is not just a trend—it's a strategic advantage.