What does the Einstein Trust Layer do?

It sits between Salesforce records and the language model. Before a prompt is sent it masks sensitive fields and grounds the request in approved data sources. After the model responds it checks the output and restores the masked values for the user. Each exchange is recorded so the interaction can be reviewed later.

Does the Einstein Trust Layer cover a model I license myself?

It governs models consumed through Salesforce's own AI infrastructure. If you connect a provider directly under a BYOM architecture, the equivalent controls have to exist on that path: masking before the callout, permission checks on what the model can read, and a stored record of each prompt and response.

How is the GPTfy Security Layer different?

It applies the same class of controls to models you license yourself. GPTfy masks across four layers before the callout, enforces Salesforce role and field-level security on what the model can read, and writes each prompt and response to a Security Audit Record. Because the provider contract is yours, retention terms are yours to set.

Does masking mean nothing is sent to the AI provider?

No, and that is worth stating plainly. Raw data stays in Salesforce, and masked data is what reaches your AI provider. Data never leaves your infrastructure, but an outbound callout does happen on every request. Any description that promises a hosted model with no outbound request at all is describing something else.

Is a trust layer enough for HIPAA or GDPR?

It is one control among several, not a certification. Masking reduces what reaches a third party, and audit records support an access review, but scope, retention terms with your provider, and your own policies all matter. GPTfy masks 16 of the 18 HIPAA PHI identifiers, since biometric identifiers and full-face photographs cannot be masked.

Einstein Trust Layer

Salesforce's LLM security abstraction: masks PII before sending to AI models, audits every prompt and response, and blocks training on customer data.

Quick answer

What is Einstein Trust Layer?

The Einstein Trust Layer is the security and governance layer Salesforce places between org data and a language model. It masks sensitive fields before the prompt is sent, grounds responses in approved sources, checks output, and records each exchange for audit. It applies to models consumed through Salesforce's own AI infrastructure.

Last updated: August 2026

The Einstein Trust Layer is Salesforce's answer to the "can we trust AI with customer data" question. It sits between Salesforce records and the underlying LLM (OpenAI, Anthropic, Google, or via BYOM). Before any prompt reaches the model, the Trust Layer: (1) masks PII like emails, phone numbers, SSNs; (2) applies dynamic grounding via approved data sources; (3) checks for toxicity; (4) logs the full transaction for audit; (5) demasks PII before returning the response to the user.

For regulated Salesforce customers in healthcare, financial services and government, a governance layer of this kind is usually what makes an AI deployment approvable at all. The controls that matter to a security reviewer are consistent: which fields are masked before the callout, what the provider is contractually allowed to retain, and whether every prompt and response can be produced on request.

The Trust Layer applies to models consumed through Salesforce's own AI infrastructure. Teams that connect a model they license directly need the equivalent controls on their side: masking before the callout, a record of each prompt and response, and permission checks on what the model is allowed to read. The GPTfy Security Layer provides that path for models brought in under BYOM.

Browse all terms

Frequently asked