What is Permission Union Review in Agentforce architecture?

In Agentforce architecture, review the agent's permissions, field-level security, sharing, and knowledge reach as one combined picture before launch. This guide explains the design decisions, controls, and implementation checks needed to apply the pattern in production.

An agent's effective access is the union of everything it can reach: object permissions, field-level security, sharing rules, flows, Apex actions, and grounded knowledge sources. Each grant may be reasonable on its own, yet the union can expose data or actions that no single reviewer ever approved together.

Reach accumulates quietly as teams add actions, topics, and integrations over time. Reviewing permissions one component at a time misses the combinations where sensitive fields, broad sharing, and a powerful action meet in a single conversation.

AI risk management cycle showing Govern, Map, Measure, and Manage with context-based risk mapping

Review the union, not the parts

  • Treat object permissions, FLS, sharing, and knowledge source access as one access surface for the agent user.
  • Apply least privilege to the union: remove any grant the agent does not need for its defined topics and actions.
  • Check grounded knowledge sources for content the agent can now surface to users who could not open those records directly.
  • Re-review the union whenever actions, topics, integrations, or sharing models change, because each change reshapes the whole.

Build the combined access picture

  1. List the agent user's permission sets and permission set groups, then map every object, field, and Apex or flow action they grant.
  2. Overlay the org's sharing model to confirm which records the agent can read and act on, not just which objects.
  3. Inventory every grounded knowledge source and verify its contents are appropriate for the agent's audience.
  4. Test with real conversations as users of different profiles, and confirm sensitive fields and records stay out of responses.
Further reading

Resources

This Tucario article is based on the following Salesforce learning and product documentation.

Continue the architecture path
TrustDefine the Agent Guardrails

Use the People, Business, Technology, and Data framework to document operational boundaries.

TrustAgentforce Trust Patterns

Understand platform protections, custom boundaries, and deliberate human handoffs.

TrustUpgrade AI with Real-World Data

Ground agents with verified knowledge sources and scoped retrievers.