Agentforce guardrails and trust patterns
Key Takeaway
Agentforce guardrails work in layers, including platform-wide policies, the Einstein Trust Layer, and the subagent instructions you write. Trust patterns are safety designs that Salesforce builds into its AI products.
Agentforce guardrails are the rules and controls that keep AI agents safe, accurate, and inside the job they were built for. Trust patterns are standard product designs that Salesforce builds into its AI products to improve safety, such as clearly disclosing AI-generated content and making handoffs to humans smooth.
The layers of guardrails
Agentforce guardrails work in these layers:
- Platform-wide policies
- Guardrails built into Agentforce and into each agent type
- Instructions you write
- The Einstein Trust Layer
- Standard Salesforce access controls