Generative AI audit trail and feedback data
Key Takeaway
The audit trail shows the masked prompt text sent to the external LLM and the full unmasked response served to the user. Audit data includes user data, but feedback data does not.
Data 360 stores two types of data for Einstein generative AI. Generative AI audit data, also called the audit trail, records how the Einstein Trust Layer handled each request, including data masking and toxicity scoring. Feedback data records how users reacted to the AI output. Both are stored in Data 360, so you can report on them. Use those reports to check that AI use meets your security, privacy, regulatory, and AI governance policies.
Audit data and feedback data
Data 360 collects two types of data for Einstein generative AI: generative AI audit data, also known as the audit trail, and feedback data. Both are stored in a Data 360 data space. You access them with Data 360 reporting and other standard platform capabilities.
| Audit data | Feedback data | |
|---|---|---|
| Comes from | LLM Gateway | Feedback API |
| Records | How the Einstein Trust Layer handled each request | How users responded to generated output |
| Includes user data | Yes | No |
Because audit data includes user data and feedback data does not, audit data is the type to review when you need to look at user data.