Grounding is the process by which an AI agent retrieves and uses specific, trusted data to ensure its responses are accurate and relevant 41 . During this process, Salesforce ' s Einstein Trust Layer performs several critical checks, including identifying Prompt Injection risks 424242 . Prompt injection refers to attempts by a user to " trick " the AI into ignoring its instructions or revealing sensitive data by entering malicious commands into the chat interface 43 . By checking for these risks during the grounding phase, the agent can ensure that the instructions it uses to generate a response remain secure and aligned with company policies 44 . While the agent respects field security, " Encrypted fields " (Option C) are handled at the data access layer rather than as part of the linguistic grounding check 45 . Standard Agentforce agents do not browse the open web for verification (Option D) as they are intended to remain grounded strictly in the company ' s verified internal data 46 .