---
title: "Audit Workday Security Groups Before Enabling AI Agents"
description: "How to find unconstrained User-Based groups, stale role assignments, and over-broad domain policies in Workday before activating delegate or ambient AI agents."
url: https://kiweely.com/guides/workday-security-audit-before-enabling-ai-agents
---

# How to Audit Over-Permissive Workday Security Groups Before Turning On AI Agents

Delegate AI agents inherit the union of an invoking user’s security groups, while ambient agents act through dedicated Ambient Agent Security Groups. Before activating any agent, audit unconstrained User-Based groups, stale role assignments, and View Security for Agent Skill pathways.

-   **Checked** against Workday on November 13
-   Applies to **Workday Security**
-   6 min read
-   Published November 13
-   By Kiweely

Before activating any Workday AI agent, audit **unconstrained User-Based security groups**, **Location Membership groups**, **stale role assignments**, and **union-of-access pathways** using **View Security for Agent Skill**, **Security Analysis for Workday Account**, and the **Role-Based Security Audit Report**. Because **delegate-mode** agents execute APIs using the union of every security group an invoking worker belongs to, and **ambient-mode** agents run autonomously through generated **Ambient Agent Security Groups (ASGs)**, any forgotten over-permission becomes immediately queryable in natural language.

For years, “security by obscurity” masked sloppy tenant configurations: an analyst who transferred out of Compensation two years ago still held a User-Based group, or a domain policy granted *View* to *All Users*, yet nobody noticed because no custom report was shared with them. Turning on [Sana Self-Service Agent](https://kiweely.com/guides/workday-sana-self-service-agent-admin-setup) or custom agents removes UI navigation obscurity overnight. The agent invokes Workday Graph, REST, SOAP, and WQL tools directly against the user’s effective domain and business process permissions.

## Why Workday’s “Union of Access” Rule Multiplies AI Agent Blast Radius

Workday’s configurable security framework evaluates target data visibility as the **union** of all security groups a user belongs to on a given securable item. If a manager belongs to a [constrained Role-Based group](https://kiweely.com/guides/workday-constrained-vs-unconstrained-security-groups) (*Manager*, scoped to their Supervisory Organization) **and** a single unconstrained group that also sits on *Worker Data: Compensation*, unconstrained access wins across the entire tenant.

Five classic configuration habits cause 90% of AI data over-exposure:

| Over-Permissive Pattern | Why It Leaks Data Through AI Agents | Delivered Audit Report / Task to Catch It | Remediation Before Agent Launch |
| --- | --- | --- | --- |
|  **Lingering User-Based Security Groups** | User-Based groups are **always unconstrained** and stay attached to the person across *Change Job* transfers forever unless manually removed. |  **View Security Group**, **Security Analysis for Workday Account**, and custom report on PBO `Security Group` (`Members`) | Convert operational roles to **Role-Based (Constrained)**; add **Remove User-Based Security Groups** service step to *Termination* and *End Contingent Worker Contract* BPs. |
|  **Location Membership Groups on Sensitive Domains** |  **Location Membership** groups are **unconstrained**: Workday does not match the member’s location to the target worker’s location. |  **View Security Groups** filtered by group type *Location Membership*; **Action Summary for Security Group** | Replace with **Organization Membership (Constrained)** using your *Location Hierarchy* (or wrap inside an [Intersection Security Group](https://kiweely.com/guides/workday-role-based-user-based-intersection-security-groups)). |
|  **Over-Broad `All Users` or `Employee As Self` Domain Grants** | Adding *All Users* (unconstrained) instead of *Employee As Self* (self-constrained) to a worker domain lets delegate agents read peer records via API tools. |  **Domain Security Policies for Functional Area** and **View Security for Securable Item** (search shortcut `secura`) | Replace *All Users* with *Employee As Self* on worker data domains; keep *All Users* strictly for metadata domains like *Graph API Descriptor*. |
|  **Stale Supervisory or Cost Center Role Assignments** | Filled positions or unrevoked inherited roles leave former leads with constrained manager access over old orgs. |  **Role-Based Security Audit Report** (`Security Outcome` = *Security Granted* / *Potential Impact*) and **Worker Roles Audit** | Revoke orphaned role assignments; verify *Job Change Vacate* entries log *Security Revoked*. |
|  **Over-Privileged Ambient Agent Security Groups (ASGs)** | Ambient agents run without a human user; granting an ASG *Modify* or *Put* on broad domains gives scheduled background runs unchecked write authority. |  **View Security for Agent Skill** and **All AI Agent Accounts** | Grant ASGs only the exact *View*/*Modify* or *Get*/*Put* permissions listed as required for that skill’s tools. |

## Step 1: Audit Skill-to-Policy Alignment with `View Security for Agent Skill`

Do not guess which Workday domains an AI agent’s tools hit under the hood. Before clicking **Activate** on any registered agent in **Agent Management Hub**, run the delivered **View Security for Agent Skill** report (requires *Manage: Agents* in *Agent System of Record*, plus *Security Administration* in *System* to run diagnostics):

1.  Prompt **View Security for Agent Skill** with your **Registered Agent** and **Agent Skills**.
2.  Inspect the **Functional Status** column (`Full Match`, `Review Required`, `Config Required`, or `Not Evaluated`), **Required Permissions** (*View* vs. *Modify*, *Get* vs. *Put*), and **Assigned Security Group** across every underlying tool.
3.  Click **Test My Setup** for each target security group. Drill down into **alternate security policy pathways** to spot cases where a user passes tool authorization not through your intended constrained group, but through an unintended unconstrained group sitting on the same domain.

Fails

```
// Auditing only the skill's "Available To" prompt in Agent Management Hub:
Skill: ESS - Human Capital Management - Base -> Available To: Employee As Self
Hidden Domain Policy Bug: "Worker Data: Current Staffing Information" also includes "US Location Membership" (Unconstrained!)
Result: Any US employee asking the agent about org staffing can retrieve records outside their own profile.
```

Works

```
// Running View Security for Agent Skill -> Test My Setup + Security Analysis for Securable Item and Account:
1. Spot "US Location Membership" as an alternate unconstrained pathway on the tool's domain policy.
2. Remove "US Location Membership" via Maintain Permissions for Security Group (or Edit Domain Security Policy Permissions).
3. Run Activate Pending Security Policy Changes and re-verify with Test My Setup.
```

## Step 2: Clean Up Sticky User-Based Groups and Put Assignments Behind Workflow

Because **User-Based** security groups follow individuals across promotions and department moves, audit every non-admin member before turning on org-wide chat agents:

1.  Run **Security Analysis for Security Group** and **Action Summary for Security Group** on high-privilege groups (*HR Administrator*, *Payroll Administrator*, *Report Writer*, *Compensation Administrator*).
2.  For sampled workers who changed departments in the past 12 months, run **Compare Security of Two Worker Accounts** (comparing the transferred worker against a clean peer in their new job profile) and **Role-Based Security Audit Report** (checking the default one-month *Effective Moment* window or wider for missing *Security Revoked* outcomes on mass actions or EIB loads).
3.  Batch-remove over-permissive groups across multiple domains in one transaction using **Maintain Permissions for Security Group** (secured to *Security Configuration*), review your staged edits via **Domain Security Policies with Pending Changes** and **Business Process Security Policies with Pending Changes**, and commit via **Activate Pending Security Policy Changes**.
4.  Lock down future User-Based sprawl: enable the **Process: User-Based Security Group Event** domain, configure business process definitions for **User-Based Security Group Event For User** and **User-Based Security Group Event for Group** so self-assignments require peer Security Admin approval, and opt in to **Disable User-Based Security Group Assignment Tasks** under **Maintain Feature Opt-Ins**.

Security Timestamps Do Not Roll Back Group Membership

**Activate Previous Security Timestamp** only reverts edits made to **domain and business process security policies** (returning later policy edits to a pending state). It **never** reverts changes to security group definitions, User-Based group assignments, or role assignments, because those take effect immediately without activation. Always export current group memberships before bulk cleanup.

## Step 3: Verify Least Privilege with Real Personas Before Enabling Agents

Remember two final verification rules before launch:

-   **Check Segregation of Duties on AI & Security Domains:** Run **Domain Security Policies for Functional Area** on *System* and *Agent System of Record* to confirm that the group editing policies (*Security Configuration* / *Setup: Agents*) is separated where required from the group activating policies (*Security Activation*), and verify **Security Exception Audit** returns zero unintended gaps.
-   **Test without Proxy where ASUs are involved:** Workday AI agents and [Agent System Users](https://kiweely.com/guides/workday-agent-passport-governance-explained) reject **Start Proxy** sessions. Combine **Test Security Group Membership** and **Security Analysis for Securable Item and Account** with non-proxy test accounts in your Implementation or Sandbox tenant to confirm a standard employee, a constrained manager, and a transferred worker each hit hard authorization boundaries on out-of-scope prompts.

Manually tracing union-of-access pathways across hundreds of domain policies, intersection groups, and 20+ agent skills takes days in spreadsheets. [Kiweely Vine](https://kiweely.com/vine) connects to your tenant, traces effective security permissions across groups and securable items, flags unconstrained leaks before you turn on AI agents, and runs [multi-persona browser verification](https://kiweely.com/guides/workday-multi-persona-proxy-testing) with explicit consent before changing a single policy ([see how consultants and HRIS teams use Vine](https://kiweely.com/vine/jobs)).

## Frequently asked questions

### Why do AI agents expose Workday security bugs that employees never noticed in the UI?

In the standard Workday UI, workers rarely guess exact report names or hidden task paths, and custom reports often apply report-level filters. Conversational AI agents call underlying REST, SOAP, Graph, and WQL tools directly, evaluating the full union of all security groups assigned to the user, so one forgotten unconstrained group exposes tenant-wide records in chat.

### Does reverting a Workday security timestamp undo accidental security group assignments?

No. Running Activate Previous Security Timestamp only reverts changes to domain and business process security policies back to a pending state. Changes to security group definitions and user-based group memberships take effect immediately without activation and are never rolled back by a timestamp revert.

### How do I keep ambient AI agents from gaining tenant-wide modify access?

When you activate an ambient skill in Agent System of Record, Workday generates a dedicated Ambient Agent Security Group (ASG). Never add that ASG to broad administrative domains; grant only the specific View, Modify, Get, or Put permissions listed in View Security for Agent Skill, or pair it in an Intersection security group where supported.

### Why do transferred employees still retain HR or Finance admin access in AI chat?

User-Based security groups are unconstrained and follow the person across job changes and transfers until manually removed. Replace broad User-Based groups with Role-Based (Constrained) groups, and add the Remove User-Based Security Groups and Terminate User Account service steps to your Termination and contract-end business processes.

## Sources checked

-   [View Security for Agent Skills and Tools](https://doc.workday.com/admin-guide/en-us/workday-ai/agent-security/view-security-for-agent-skills-and-tools.html)
-   [Configure Security Policies for Agent Skills](https://doc.workday.com/admin-guide/en-us/workday-ai/agent-security/configure-security-policies-for-agents.html)
-   [Security Group Configuration and Constraints](https://doc.workday.com/workday-education/en-us/course-manuals/security-for-administrators/security-group-configuration-and-constraints.html)
-   [Choosing the Right Security Group Type and Delivered Security Reports](https://doc.workday.com/workday-education/en-us/course-manuals/security-for-administrators/choosing-the-right-security-group-type.html)
-   [Set Up Role-Based Security Reporting](https://doc.workday.com/admin-guide/en-us/manage-workday/roles/set-up-role-based-security-reporting.html)

## Related guides

[All guides](https://kiweely.com/guides)

-   [### What Is Workday Agent Passport and How Do Admins Audit or Stop an AI Agent? Workday Agent Passport shows verified security stamps, assessors, and grant dates on each agent profile in Agent Management Hub. To audit or stop an AI agent, inspect View AI Agent User Audit Trail and click Deactivate Agent or disable its Agent System User OAuth 2.0 client.](https://kiweely.com/guides/workday-agent-passport-governance-explained)
-   [### What Workday Admins Need to Configure Before Turning On Sana Self-Service Agent Before activating Sana Self-Service Agent, run Set Up AI Admin Security Permissions, enable Agent System of Record, and add All Users to Graph API Descriptor. Keep the Guide skill enabled and align every skill’s Available To groups with underlying tool security policies.](https://kiweely.com/guides/workday-sana-self-service-agent-admin-setup)
-   [### Workday Role-Based, User-Based, Intersection, and Aggregation Security Groups User-Based security groups grant unconstrained tenant-wide access to specific accounts, while Role-Based groups constrain access by organizational assignment. Intersection groups combine groups with AND logic to narrow access, and Aggregation groups combine groups with OR logic.](https://kiweely.com/guides/workday-role-based-user-based-intersection-security-groups)
-   [### Workday Domain Security vs. Business Process Security Policies: Which Controls What A domain security policy controls View/Modify (and Get/Put) access to reports, report fields, setup tasks and web services. A business process security policy controls who can start, approve, cancel, rescind or view a transaction. Both need activation.](https://kiweely.com/guides/workday-domain-vs-business-process-security-policies)
-   [### Constrained vs unconstrained security groups in Workday: which workers can a member see? An unconstrained group gives members access to every worker in the tenant for the secured item. A constrained group limits them to the organizations they are assigned to, such as an HR Partner for one supervisory organization.](https://kiweely.com/guides/workday-constrained-vs-unconstrained-security-groups)

### More guides on Custom agents and Workday AI

[All 91 guides](https://kiweely.com/guides#custom-agents)

-   [Calling External Glean and Sana Agents From Workday Extend Without Hitting 408 Request Timeouts](https://kiweely.com/guides/workday-extend-external-agents-glean-sana-408-timeout)
-   [AI Gateway Document Intelligence (/ml/docint/v1/ids): Converting Workday Drive Base64 to Binary, Resume Layout Pitfalls, and Fixing Opt-In Errors](https://kiweely.com/guides/workday-document-intelligence-base64-binary-opt-in-errors)
-   [Why a Workday Custom Agent Has No Credentials Tab in Agent Management Hub, and How to Expose It Through A2A](https://kiweely.com/guides/workday-custom-agent-missing-credentials-tab-a2a)
-   [Passing Large CSV File Attachments From a Workday Custom Agent Flow to Orchestrate Without Hitting Context Limits](https://kiweely.com/guides/workday-custom-agent-csv-file-attachment-orchestrate)
-   [Preventing Duplicate Initiate\_Change\_Job Business Process Events When a Workday Custom Agent Calls a Write Tool](https://kiweely.com/guides/workday-custom-agent-duplicate-initiate-change-job)
-   [Why Custom Agent Actions Fail With S22 Permission Denied or Task Not Authorized in Agent Chat](https://kiweely.com/guides/workday-custom-agent-actions-permission-denied-tool-wids)
-   [Fixing invalid\_request OAuth Token Exchange Errors When Connecting Microsoft Copilot Studio to Workday A2A](https://kiweely.com/guides/workday-a2a-copilot-studio-oauth-invalid-request)
-   [Why Custom Agent Skills Deployed From the Developer Site Don’t Appear in Agent Management Hub, and ASOR Tenant Prerequisites](https://kiweely.com/guides/workday-agent-management-hub-skills-not-showing)

### From the blog on Custom agents and Workday AI

[All posts](https://kiweely.com/blog)

-   [Will AI Replace Workday Consultants? No. The Agent Works for You](https://kiweely.com/blog/will-ai-replace-workday-consultants)
-   [ServiceNow AI agents vs Workday AI agents: operations vs HCM/finance build](https://kiweely.com/blog/servicenow-ai-agents-vs-workday-ai-agents)
-   [How Workday prospects should run an AI-agent-driven implementation methodology](https://kiweely.com/blog/workday-prospect-ai-agent-implementation-methodology)

## Let Workday check the work.

Kiweely writes WQL and builds orchestrations in your own tenant, and Workday, not Kiweely, says whether they passed.

[Start for free](https://kiweely.com/download) [See how it works](https://kiweely.com/vine/how-it-works)

## Get the next guide in your inbox.

One email a month with new guides and one Extend tip. Nothing else.
