You are here:
Einstein Audit, Analytics, and Monitoring Setup Control
Centralizes the logging of Generative AI prompts, LLM responses, and user feedback into Data Cloud for compliance, auditing, and model optimization.
Control Name
Einstein Trust Layer - Audit and Feedback Data
Control Overview
Centralizes the logging of Generative AI prompts, LLM responses, and user feedback into Data Cloud for compliance, auditing, and model optimization.
Description
Enables the permanent storage of Trust Layer signals and the user feedback, moving data from transient memory into persistent data model objects (DMOs); allowing administrators to monitor the performance and accuracy of the agents.
Recommended Configuration
Go to Einstein Audit, Analytics, and Monitoring Setup. Enable "Collect and Store Einstein Generative AI Audit Data" and "Capture User Feedback". Make sure that Data Cloud is fully provisioned to host these data streams.
Security Impact
Provides the forensic trail necessary to investigate accuracy, performance, data leaks, or toxic interactions, to make sure that the company meets its requirements based on its use cases.
Business Impact
Creates a central repository of logs for your agents where real-world user feedback and data are used to refine prompt templates and agent instructions.
Security Risk If Not Configured
If an agent provides incorrect or harmful advice, there is no permanent record to diagnose the cause or prove what was said.
Threat Scenarios
Accountability Gap: Absence of logs to verify the interaction in case of conflicts. Blind Spots: Systematic bias in AI responses goes undetected for months.
Estimated CVSS Score Range
Critical (9.0–10.0).
Risk Impact Considerations
Enabling this configuration captures the relevant audit and feedback data. However, customers should implement appropriate alerting and monitoring controls to make sure that relevant anomalies are reported, and false positives are addressed to reduce noise.
Higher Risk When
Audit data collection is not enabled to save on Data Cloud credits, or when feedback is collected but never reviewed by a human-in-the-loop.
Low Risk When
Customers have implemented customized audit logging procedures where agent interactions are logged for prompt injection, toxicity detection, and other capabilities to monitor the performance and accuracy.
Business and Integration Considerations
Storage Costs: This data consumes Data Cloud credits. Latency: Audit data can take up to 24 hours to appear in reports.
Security Health Review Guidance
Security Health Review verifies the "Enabled" status of the Audit Data toggle and flags instances where feedback loops are inactive, preventing the company from improving its AI safety.
Who Is Impacted
Data governance officers, AI admins, customer service managers, and legal/compliance auditors.
