LLM data masking for agents versus embedded generative AI features
Key Takeaway
Data masking in the Einstein Trust Layer works for embedded generative AI features, such as Einstein Service Replies, but it is disabled for agents. Agents rely on other Einstein Trust Layer protections, such as zero data retention.
The Einstein Trust Layer can detect sensitive data, such as personally identifiable information (PII), and mask it before a prompt reaches a large language model (LLM). Embedded generative AI features, such as Einstein Service Replies and Einstein Work Summaries, use this masking. Agents don't use it, because masked data can remove the context an agent needs to give a useful answer. Instead, agents rely on other Einstein Trust Layer protections, such as zero data retention.
What LLM data masking is and where you set it up
LLM data masking is an Einstein Trust Layer feature that detects sensitive data and masks it, so the LLM doesn't see it.
You set up data masking on the Einstein Trust Layer setup page. The page has two tabs: one for LLM data masking and one for content safety and security.
Before you can set up the Einstein Trust Layer, you must enable Einstein Generative AI and configure Data 360. The Einstein Trust Layer needs Data 360 to work correctly and protect your data.
Data masking is enabled by default.