---
title: What permissions does a Salesforce Agentforce agent run with?
url: https://opencontroller.lyzr.com/blog/agentforce-agent-user-permissions
published: 2026-11-14
updated: 2026-11-14
author: Lyzr
---

# What permissions does a Salesforce Agentforce agent run with?

**It depends on the agent type. Employee agents run as the signed-in user and follow that user's permissions and org security settings. Service agents, which answer customers, run as a dedicated agent user: a Salesforce user with the Einstein Agent User profile and Agentforce permission sets, no password and no login. Whatever that agent user can read, any customer talking to the agent can potentially reach.**

## How do employee and service agents differ?

| | Employee agent | Service agent |
|---|---|---|
| Who uses it | Internal staff in Salesforce | Customers and external users on channels |
| Runs as | The signed-in user | A dedicated agent user |
| Data visibility | The user's own permissions, sharing and org security | The agent user's permission sets and sharing |
| Main risk | Over-privileged employees get a faster way to reach data | One broad agent user exposes data to every customer conversation |
| Audit attribution | Actions appear under the employee | Actions appear under the agent user |

Salesforce's Trailhead material describes employee agents as logged in as the person they are helping, so a manager and an employee get different results from the same agent.

## What is the Agentforce agent user?

Salesforce CLI documentation for `org create agent-user` describes the default agent user:

- Profile: **Einstein Agent User**
- Required permission sets: **AgentforceServiceAgentBase**, **AgentforceServiceAgentUser** and **EinsteinGPTPromptTemplateUser**
- No password; it cannot log in to Salesforce
- Referenced by the agent's Agent Script through the `default_agent_user` setting
- Extra access is added later by assigning more permission sets or permission set licences

Those base permission sets let the agent run. They do not decide what data it can see. That comes from the permission sets you add for each topic and action.

## Where do agents get more access than intended?

Three patterns cause most over-exposure.

1. **Broad object permissions on the agent user.** "View All" or "Modify All" on objects such as Case or Opportunity bypass sharing entirely.
2. **Actions running in system context.** Flows set to run in system context, and Apex that does not enforce sharing or user mode, can read records the running user cannot.
3. **One agent user for many agents.** Each new agent adds permission sets to a shared user, and the union grows quietly.

## How should you design permission sets for Agentforce?

Use this checklist for every service agent.

1. Create **one agent user per agent**, not one per org, so access and audit stay separable.
2. Start from the three base permission sets only.
3. Add one permission set per topic, named after the topic, with only the objects and fields its actions need.
4. Never grant View All or Modify All to an agent user.
5. Rely on sharing rules to limit records, and test with a record the agent should not see.
6. Set flows used as agent actions to run in user context where possible. In Apex, enforce sharing and use user-mode database operations.
7. For employee agents, review the riskiest user profiles first; the agent reflects their access.
8. Recertify agent user permissions on the same cycle as other non-human identities. See [access reviews for AI agents](/blog/access-reviews-for-ai-agents).

## What does this mean for audit?

For service agents, the agent user is the actor in Salesforce logs, so record which customer session triggered each action. Salesforce provides the setup audit trail for configuration changes and Event Monitoring for activity; session-level traces are covered in [how to audit Agentforce sessions with session tracing](/blog/audit-agentforce-sessions-with-session-tracing). Standard logs show what the agent did, not on whose behalf, so correlating session and agent user is your job.

For employee agents, actions appear under the employee. That is better for attribution but makes it harder to tell agent actions from manual ones, so tag or filter by channel.

More platform guidance is in [how to govern Salesforce Agentforce agents](/blog/govern-salesforce-agentforce-agents).

## How Opencontroller does this

Opencontroller connects to Salesforce as a third-party agent platform, so Agentforce agents appear in the same governed registry as agents on AWS, Azure, Google Cloud and other platforms, with owners and versions. Posture detects risky permissions and configurations, such as broad grants on an agent's identity. Per-agent Block and Quarantine act through the connected platform. See the [AI agent control plane explainer](/blog/what-is-an-ai-agent-control-plane) or [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Does an Agentforce employee agent see more than the user?**
It is designed not to. Employee agents run as the signed-in user and follow that user's permissions and org security settings, but actions running in system context can bypass sharing.

**Can someone log in as the Agentforce agent user?**
No. The default agent user has no password and cannot log in. Only admins with the right permissions can view or edit it in Setup.

**Should several service agents share one agent user?**
Avoid it. Separate agent users keep permissions minimal and make audit logs attributable to a single agent.

## Key facts

Salesforce Agentforce employee agents run as the signed-in user and follow their permissions and org security. Service agents run as a dedicated agent user with the Einstein Agent User profile and the AgentforceServiceAgentBase, AgentforceServiceAgentUser and EinsteinGPTPromptTemplateUser permission sets; it has no password and cannot log in, and is set via default_agent_user in Agent Script. Main risks: View All/Modify All on agent users, system-context flows and Apex, shared agent users. Opencontroller connects to Salesforce and Posture detects risky permissions. Pricing starts from $250K per deployment per year.

## Related

- [How do you govern Salesforce Agentforce agents?](https://opencontroller.lyzr.com/blog/govern-salesforce-agentforce-agents.md)
- [How do you audit Agentforce agent sessions with Session Tracing?](https://opencontroller.lyzr.com/blog/audit-agentforce-sessions-with-session-tracing.md)
- [How do you run access reviews for AI agents?](https://opencontroller.lyzr.com/blog/access-reviews-for-ai-agents.md)
- [More on Platform Guides](https://opencontroller.lyzr.com/blog/cluster/platform-guides)
