Skip to content
KiweelyKiweelyKiweely

Review security group changes an AI assistant suggested in Workday

Deployment Agent can name a domain and a missing View group. As documented, you still edit the policy (change-making is in partner early access), activate pending changes, and confirm View Security for Agent Skill is not Config Required. Constrained groups cannot administer Agent System of Record.

  • Checked against Workday on
  • Applies to Workday Extend
  • 6 min read
  • Published
  • By Kiweely
On this page

Treat an AI security recommendation as a list of objects to open, not as a change that is already safe. Workday’s Ask-skill example asks Deployment Agent which domain controls the Bonus Plan Override field, which groups have View, and how to update the policy. The sample answer names Worker Data: Compensation Explicit and a missing Payroll Specialist (role-based) group, then points you at editing domain permissions and activating pending changes. As documented, you do those tasks and the chat does not activate them. With Configuration Cloud, in opt-in early access for Workday partners since 18 Nov 2026, Deployment Agent can carry out changes once approved.

Review in a non-production tenant. Deployment Agent is not a Production assistant. Agent System of Record (ASOR) security is configured per tenant and does not migrate.

Sort the suggestion before you edit anything

The reply tells you to change Legal target Reject or redirect
The reply tells you to changeA domain’s report/task permissions (View or Modify) Legal targetEdit Security Policy Permissions on that domain, then Activate Pending Security Policy Changes Reject or redirectA group that does not exist. A domain you cannot find.
The reply tells you to changeWho may open Deployment Agent Legal targetThe group must be on Deployment Agent Users and on the skill’s Available To. Workday validates this and blocks groups that are not on the domain. Reject or redirectPutting only Available To and skipping the domain.
The reply tells you to changeWho may administer agents Legal targetMaintain Permissions for Security Group on ASOR policies. Unconstrained groups only. Reject or redirectA constrained HR Partner role as the ASOR admin group. The setup topic does not allow it.
The reply tells you to changeWho may use a Self-Service skill Legal targetAvailable To on that skill. Guide must stay enabled or the agent does not run. Reject or redirectA tool-level security group. Tools inherit the skill’s groups. You cannot assign a tool to one group.
The reply tells you to changeIntegration Get or Put Legal targetOnly if you are looking at a web service policy and you meant to grant it. The Ask example’s result is View on a worker data domain, not an integration permission. Reject or redirectAdding Modify because the reply said “access.” View and Modify are different. Modify includes what View includes.

View Security for Agent Skill is how you review the third case when the change is about an agent. You need Manage: Agents. Prompts: Registered Agent, Agent Skills.

Status Review rule
StatusFull Match Review ruleThe Available To group has every required policy for the tool. Still ask whether that group should.
StatusReview Required Review rulePartial required policies. Do not activate the agent on this status.
StatusConfig Required Review ruleA required policy has no groups. Fix these first.
StatusNot Evaluated Review ruleAvailable To is empty. Nobody has been tested.

Required Permissions show View or Modify. Additional Security (Optional) is extra field access, not the minimum for the tool. Test My Setup (Security Administration) shows whether a group can reach the tool and which other policies also grant it.

Self-Service Agent evaluates both access to the agent and access to the tools’ APIs. Align Available To with the groups that can run the tools. They do not have to be an identical list, but a person needs to pass both checks. The setup topic tells you to compare them.

The review sequence

  1. Paste the suggestion into the change ticket with the tenant name and the prompt you used. Include “in my tenant” prompts only if that phrase was actually sent. Without it, the answer may be generic.
  2. Open the named domain. If the domain name does not exist, stop. Do not pick a similarly named domain.
  3. Read current View and Modify rows. Note groups the reply did not mention. You are adding, not replacing, unless the reply explicitly says to remove a group and you agree.
  4. For a Deployment Agent chat group, confirm the group is on Deployment Agent Users before you put it on Available To.
  5. For an ASOR admin change, confirm the group is unconstrained. Grant View and Modify only on the policies you intend: Agent Compliance, Agent Management Hub, Manage Agents, Reports: Agent Reporting, Reports: AI Agent Security, Setup: Agents.
  6. For an agent skill, run View Security for Agent Skill. Clear Config Required. Record Review Required groups you refuse to widen.
  7. Edit the policy. Prefer a new row when the new group needs View and the existing row is Modify, so you do not copy the stronger permission by accident.
  8. Activate Pending Security Policy Changes, with a comment that names the domain and the group.
  9. Test My Setup or a proxy session as a user in the group and as a user outside it. Pending changes that you forgot to activate will fail this test. That failure is real.
  10. Thumbs up or thumbs down on the Deployment Agent answer after you know whether the domain was right. The rating does not roll back security.

Customer Central can be asked how the same change would affect another tenant. The published follow-up names the other tenant. Do not assume impl2 was inspected because you typed “downstream.”

Self-Service tool messages are not security

If the suggestion says “restrict bonus history with a message,” that is not a domain change. The configure-tools topic lets you disable a tool (unless the Enabled toggle is unavailable) and supply a disabled message. The agent writes its own sentence from that message. Skill and tool changes can take up to 10 minutes to cache, and the tools topic says a person already in a conversation may not see the change until a new conversation. Security groups on the skill are unchanged when you only edit the message.

Disabling one tool does not change subscription or skill-level billing for the tools that stay on.

Failure modes

Result Likely miss
ResultValidation error on Deployment Agent Available To Likely missGroup is not on Deployment Agent Users.
ResultCannot select the security group on an ASOR policy Likely missGroup is constrained.
ResultUser still cannot see the field Likely missActivation did not run, the group membership is wrong, or the field is secured by more than one domain.
ResultUser can see the field but the agent tool fails Likely missAvailable To passed and the tool policy did not, or the status was Review Required.
ResultA whole organization gained Modify Likely missThe new group was added to an existing Modify row, or you used an unconstrained group for a worker-facing skill.
ResultSandbox review does not match Production Likely missASOR and domain edits do not migrate by themselves.

Frequently asked questions

Should we accept the domain name in the example as our domain?

Only if your tenant uses that field and the policy you open is Worker Data: Compensation Explicit. The example is a sample result for Bonus Plan Override. Your field may be a different domain. The prompt is what you reuse. The sample domain is what you verify.

Do interaction groups and API groups have to match person for person?

The Self-Service setup says to evaluate alignment. A person must be allowed on the skill and allowed on the tool’s domain or business process policy. The lists can differ. View Security for Agent Skill is the check, not a spreadsheet comparison done from memory.

Can the AI remove a group?

As documented for Ask, only if you remove it in Edit Security Policy Permissions and activate. Configuration Cloud, in partner early access, can carry out approved changes. Read the remaining rows before you confirm. There is no published undo inside the chat.

What do we file if Test My Setup disagrees with the reply?

The report status, the tool name, the Available To group, and whether you had already activated. That packet is the review. The chat is the proposal.

Sources checked

Frontmatter pages, fetched 2026-11-16.

Sources checked

Related guides

All guides

Let Workday check the work.

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

Get the next guide in your inbox.

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

We’ll email you once to confirm. Unsubscribe in one click from any issue.