Prompt Engineering20% of the exam
Einstein Trust Layer security and privacy
What each Einstein Trust Layer feature does on the prompt and response journey (secure data retrieval, data masking, prompt defense, the TLS-secured LLM gateway, zero data retention, toxicity scoring, demasking and the audit trail in Data 360), and the exam's favourite exception: data masking is disabled for agents.
Identify the security and privacy features of the Trust Layer.
~8 min read · 9 practice questions ·
What the exam expects
- Trace a prompt through the Trust Layer and name what happens at each stage, out to the model and back.
- Match a security or compliance concern to the feature that addresses it: zero data retention, data masking, prompt defense, the LLM gateway, toxicity scoring or the audit trail.
- Explain how masking detects sensitive data (pattern-based, machine learning, field-based) and how demasking restores it.
- Recognise where the Trust Layer stops: masking is disabled for agents, it applies only to generative AI and Agentforce features, and it does not replace field-level security or Shield encryption.
- Key terms: Einstein Trust Layer (Salesforce's set of agreements, security technology and data and privacy controls around generative AI); zero data retention (providers keep nothing and train on nothing); data masking (replacing sensitive values with placeholders before the prompt leaves); prompt defense (system policies added to the prompt); LLM gateway (the single governed, TLS-encrypted channel to model providers); audit trail (prompts, responses and trust signals stored in Data 360).
Key facts
Scope
- Detail
- Trust Layer capabilities apply only to generative AI and Agentforce features
Masking and agents
- Detail
- Data masking for LLMs is disabled for agents
Masking default
- Detail
- Enabled by default; entities or categories can be turned on or off
Masking detection
- Detail
- Pattern-based (regex, context words, ML) and field-based (Shield or data classification metadata)
Context window with masking
- Detail
- Limited to 65,536 tokens
Zero data retention
- Detail
- Not retained, not used for training, deleted after the response returns
Gateway
- Detail
- Governs provider interactions; TLS encryption in transit
Toxicity score
- Detail
- Five categories; overall score 0 to 1, 1 most toxic
Audit storage
- Detail
- Prompts, responses, trust signals and feedback in Data 360, with prebuilt reports
Audit prerequisite
- Detail
- Turn on generative AI data collection and storage; install the report package
Sources for this section (6)
- Designed for Trusthelp.salesforce.com (opens in a new tab)
- Set up Einstein Trust Layerhelp.salesforce.com (opens in a new tab)
- Supported Modelsdeveloper.salesforce.com (opens in a new tab)
- Einstein Trust Layerhelp.salesforce.com (opens in a new tab)
- Follow the Response Journeytrailhead.salesforce.com (opens in a new tab)
- Audit Trailhelp.salesforce.com (opens in a new tab)
- Why can one user's summary omit a field another user's includes?
- When does Trust Layer data masking not protect a prompt?
- Which two methods detect data to mask?
- Where are toxicity scores stored?
The questions are free to read. Answers with their sources and self-grading are included with the pack.
Keep reading with the Agentforce Specialist pack
- Core concepts
- Exam traps
- Worked scenario
The pack includes:
- All 26 lessons of the Agentforce Specialist guide, with reading progress
- Interactive decision walks and step-by-step sequence steppers
- Inline practice questions and one-click topic practice (9 questions for this lesson)
- An official source cited for every claim
- An AI tutor that answers from the guide
- Unlimited timed practice exams