Salesforce · Infrastruttura · YoctoIT tech page

Trust Layer

The guardrail of Salesforce AI: grounding on company data, masking and audit — powerful AI inside clear rules.

FOCUS · AI WITH RULESZero retention, masked data and an audit trail: the GenAI the DPO approves
Request the Free Assessment Learn more
YoctoIT material for clients and partners · Salesforce, Agentforce, Tableau, Slack and the other products mentioned are trademarks of Salesforce, Inc.
01 · What it is

Einstein Trust Layer, made clear.

The Trust Layer sits between Salesforce and the models: it anchors the answers to YOUR CRM data (grounding), masks sensitive data before it reaches the LLM, imposes zero retention on the providers, filters toxicity and records everything in an audit trail. It's what makes Einstein and Agentforce adoptable in the company, not just demoable.

Zero retention
the prompts don't train the models: contracts and architecture guarantee it
Masking
sensitive data replaced before the LLM, restored after
Audit
every prompt and answer tracked: the AI that answers the auditor too
Einstein Trust Layer
OFFICIAL SALESFORCE BRANDING · TRUST LAYER
SCHEMA UFFICIALE · PRINCIPI TRUSTED AI · FONTE: SALESFORCE
OFFICIAL DIAGRAM · TRUSTED AI PRINCIPLES · SOURCE: SALESFORCE
02 · How to use it well

The things that make the difference.

The prompt's journey

The user or the agent asksin natural language
CRM grounding
PII masking
Policies & filters
context · privacy · control
LLM (multi-model gateway)the engine, interchangeable
Audit trail & feedbackeverything recorded, everything improvable
AI in the company, with a seatbelt

Grounding on the right data

The AI answers from the CRM and the knowledge: data quality returns center stage.

Permissions respected

The agent sees only what the user would see: the model doesn't bypass the profiles.

Policies per use case

Free drafts, actions with approval: the risk calibrated on the task.

The dossier for the DPO

Architecture, flows and guarantees documented: the adoption that passes the committee.

03 · In depth

The Trust Layer taken apart: what it really does

The Einstein Trust Layer sits between your data and the LLMs: the secure gateway routes to the models (BYO-LLM too), grounding injects the CRM/Data Cloud context, data masking replaces PII before sending (and restores it after), zero retention is contractual with the providers (the prompts train nothing), toxicity detection evaluates the answers, the audit trail records everything in Data Cloud for the DPO and compliance.

  • Data masking — PII tokenized before the LLM: the model never sees it
  • Zero retention — contracts with the providers: prompts and outputs don't train
  • Grounding — the context from YOUR records: relevant answers, not generic ones
  • BYO-LLM — OpenAI, Anthropic, your own on Bedrock/Vertex: the model is a choice
  • Toxicity scoring — the answers evaluated before reaching the user
  • Audit trail — every prompt tracked in Data Cloud: the evidence for the DPO
04 · Numbers and lifecycle

The numbers that matter.

0
retention of the prompts at the providers
100%
of the AI interactions audited
PII
masked automatically: GDPR in the path
multi
LLM: independence from the single vendor
Enterprise AI is judged by its guardrails: policies, masking and audit configured by us — the AI that passes legal's and the DPO's scrutiny.
05 · Use cases

Where it really pays off.

GenAI in service

Answers suggested, grounded and tracked.

Assisted selling

Emails and summaries generated from the opportunity's real data.

AI compliance

The AI Act faced with the evidence ready.

AI without guardrails stays in demo: with the Trust Layer it goes to production — configured well.