---
title: "Submitting Workday Time Off Safely from Claude & ChatGPT"
description: "Step-by-step architecture of submitting Workday PTO from Claude or ChatGPT: live balance checks, validTimeOffDates, out-of-band browser approval, and worker…"
url: https://kiweely.com/blog/submitting-workday-time-off-from-claude-chatgpt-with-approval
---

# How Submitting Workday Time Off from Claude or ChatGPT Works Safely End to End

Submitting Workday time off from Claude, ChatGPT, or Gemini is only safe when reads run under the employee’s own Workday OAuth grant, Workday validates eligible dates first, and the final POST pauses for an out-of-band browser approval card that prompt injection cannot click.

-   By Kiweely
-   November 15
-   6 min read

Letting an employee type *“Take Thursday and Friday off next week as PTO”* in Claude, ChatGPT, or Gemini is only enterprise-safe when four architectural controls work together: user-delegated Workday OAuth (never a shared Integration System User), Workday’s own multi-step eligibility and calendar validation, an **out-of-band browser approval card** showing *Now* versus *After* outside the chat window, and native execution under the worker’s own Workday audit trail.

Most proof-of-concept AI demos skip straight to `POST /requestTimeOff` with a service account and treat a conversational *“Are you sure?”* message from the model as a security boundary. In a real Workday tenant, that shortcut breaks hourly vs. daily plan rules, submits 8 hours on company holidays, attributes the transaction to an integration user, and leaves write endpoints exposed to indirect prompt injection.

Here is how a governed end-to-end time-off submission actually works against Workday’s **Absence Management** (`absenceManagement/v5`) and **Business Process** (`businessProcess/v1`) APIs—and how **[Kiweely Bridge](https://kiweely.com/bridge)** (coming next) enforces every step.

## The Four-Stage Lifecycle of a Safe Conversational Time-Off Request

Trace what happens from the moment an employee asks *“Can I take November 24 and 25 off using Vacation, and submit it if I have enough hours?”* to the moment their manager sees the approval in Workday **My Tasks**:

| Stage | Workday Endpoint / Mechanism | What Happens | Security & Governance Guarantee |
| --- | --- | --- | --- |
|  **1\. Eligibility & Balance Read** | `GET /absenceManagement/v5/balances` & `GET /workers/{ID}/eligibleAbsenceTypes` | Reads live accrued balance as of the requested dates and checks eligible Time Off Types (and eligible Positions if multi-job). | Uses employee’s sealed user-delegated OAuth token; honors *Absence Balance Visibility* and plan eligibility rules. |
|  **2\. Calendar & DTI Validation** | `GET /absenceManagement/v5/workers/{ID}/validTimeOffDates` | Evaluates the worker’s work schedule, holiday calendar, Days To Include (DTI), and default daily quantity (8 hours vs. 1 day). | Prevents requesting PTO on non-working days or holidays before any write is staged. |
|  **3\. Out-of-Band Browser Check** | Pending action hash + Browser Approval Card (*Now* vs. *After*) | Connector pauses the write, returns an approval URL, and renders current vs. post-request balances and dates in the browser. | Model cannot click **Allow**; prompt injection inside chat cannot forge browser session confirmation. |
|  **4\. Business Process Submission** | `POST /absenceManagement/v5/workers/{ID}/requestTimeOff` | Submits the `Request Time Off` event with `action.id` = `d9e4223e446c11de98360015c5e6daf6` (*Submitted*), then reads `/events/{ID}/inProgressSteps`. | Audited in Workday under the employee’s own account; routes to the real manager in **My Tasks**. |

## Step 1 & Step 2: Why Workday Requires Multi-Step Pre-Validation Before Any Write

In Workday, you cannot safely guess a time-off payload from natural language alone. Absence plans vary across exempt, non-exempt, part-time, and global populations:

-   Some plans track **hours** (`Daily Quantity` = `8`), while others track **days** (`Daily Quantity` = `1`), or calculate quantity dynamically from required `Start Time` and `End Time` fields.
-   Workers holding multiple positions (for example, a 75% FTE primary role and a 25% FTE secondary role) may have **Require Position on Time Off Request** enabled on a worker-based plan where only positions meeting `FTE >= 50%` are eligible.
-   If a worker asks for Wednesday through Friday during Thanksgiving week, Thursday and Friday may already sit on their Workday **Holiday Calendar** or fall on unscheduled days excluded by **Restrict Requests to Days to Include on Calendar**.

Workday’s `absenceManagement/v5` REST API solves this with a strict three-call read sequence before any write occurs:

1.  **Check current and projected balances:** Query `GET /balances` (and `GET /common/v1/workers/{ID}/timeOffPlans`) to verify accrued hours/days as of the requested dates—not just today’s balance, which omits future accruals or already-scheduled December PTO.
2.  **Fetch eligible absence types:** Call `GET /workers/{workerID}/eligibleAbsenceTypes` to retrieve the exact Workday ID for *Vacation* or *PTO* that the employee is eligible for on the first requested date.
3.  **Validate the specific dates:** Call `GET /workers/{workerID}/validTimeOffDates?timeOff={timeOffTypeID}&date=2026-11-24&date=2026-11-25`. Workday returns which dates are valid working days, the unit of time, and the default daily quantity.

If November 26 is a company holiday, the assistant tells the employee immediately in chat: *“November 26 is already a scheduled company holiday on your calendar, so you only need 16 hours across November 24 and November 25. Your Vacation balance will go from 64 hours to 48 hours.”*

## Step 3: Why In-Chat Confirmation Fails—and How Out-of-Band Browser Approval Works

Suppose an admin enables the *Request Time Off* write tool. Why shouldn’t the assistant simply ask *“Reply confirm to submit”* inside the chat thread?

If an employee pastes an untrusted email thread, meeting transcript, or document into Claude or ChatGPT while Workday tools are connected, hidden instructions inside that text (indirect prompt injection) can steer the model into calling an enabled write tool—or calling a second `confirm_action` MCP tool—without the user intending to mutate Workday. Even without an attacker, LLMs occasionally hallucinate date offsets (“next Friday” resolved to the wrong week) when constructing JSON arguments.

**[Kiweely Bridge](https://kiweely.com/bridge/security)** eliminates chat-only writes structurally:

1.  **Read-only by default:** Write tools do not even appear in the assistant’s MCP tool list unless your admin explicitly enables that specific task in Bridge.
2.  **Short-lived pending hash:** When the assistant calls an enabled write tool, Bridge **does not** execute `POST /requestTimeOff` against Workday. Instead, Bridge validates the parameters, stores only a short-lived cryptographic hash of the pending action, and pauses.
3.  **Out-of-band browser approval card (*Now* vs. *After*):** The employee opens a dedicated browser approval card authenticated to their session—completely outside the LLM’s tool interface. The card displays deterministic Workday facts side by side:

text

```
Workday Action Confirmation — Request Time Off
Worker:          Maya Lin (Signed in via Workday OAuth)
Absence Type:    Vacation (Paid Time Off Plan)
Dates Requested: Tue Nov 24, 2026 (8.0 hrs)
                 Wed Nov 25, 2026 (8.0 hrs)

Balance Now:     64.00 hours (as of Nov 23, 2026)
Balance After:   48.00 hours (projected Nov 26, 2026)
Next Approver:   Approval by Manager (David Chen)

[ Deny ]                                     [ Allow in Workday ]
```

Prompt injection inside a chat window cannot click a button in the employee’s authenticated browser tab, and date hallucinations are caught immediately because the human sees exact calendar dates and *Now* vs. *After* balances before Workday ever receives the `POST`.

## Step 4: Executing `POST /requestTimeOff` Under the Worker’s Own Audit Trail

Only after the employee clicks **Allow** in the browser does the connector send the write payload to Workday using that employee’s decrypted-in-memory user-delegated OAuth grant:

json

```
{
  "businessProcessParameters": {
    "action": {
      "id": "d9e4223e446c11de98360015c5e6daf6"
    },
    "comment": "Taking Tuesday and Wednesday before the holiday"
  },
  "days": [
    {
      "date": "2026-11-24",
      "dailyQuantity": 8,
      "timeOffType": {
        "id": "a18c426056a1100018b52d4e99c10042"
      }
    },
    {
      "date": "2026-11-25",
      "dailyQuantity": 8,
      "timeOffType": {
        "id": "a18c426056a1100018b52d4e99c10042"
      }
    }
  ]
}
```

Because the token was issued via **Register API Client** (Authorization Code Grant with [PKCE](https://kiweely.com/guides/workday-api-client-pkce-sso-without-workday-password)) rather than an Integration System User:

-   Workday enforces both the **FLW Service** domain and the [Request Time Off business process security policy](https://kiweely.com/guides/workday-domain-vs-business-process-security-policies) (*Employee as Self*).
-   Workday’s **Process History** tab records `Maya Lin`—never `ISU_Chatbot_Prod`—as the event initiator.
-   Any tenant custom validations or condition rules configured on *Request Time Off* evaluate identically to the Workday web UI.
-   Follow-up read calls to `GET /businessProcess/v1/events/{eventID}/inProgressSteps` confirm the exact manager or timekeeper now holding the approval step in **My Tasks**, and stream that confirmation back to Claude, ChatGPT, or Gemini.

Zero Stored Payloads After the Request Completes

Throughout all four steps, Kiweely Bridge writes zero absence records, balances, worker names, or comments to disk, databases, or logs. Structured telemetry records only timestamp, tenant ID, tool name, HTTP status, and latency.

## Governance Without Per-Action Credit Burn

**Kiweely Bridge** ([coming next](https://kiweely.com/bridge)) brings this exact user-delegated read and browser-verified write workflow to **Claude, ChatGPT, and Gemini** on any standard Workday tenant—requiring no Workday AI agreement, no Agent System of Record SKU, and no per-call Flex Credits ([compare Bridge against Workday Agent Gateway, StackOne, and in-house builds](https://kiweely.com/bridge/compare)).

Bridge is priced flat per company by licensed Workday employee count with unlimited calls across Production and all Sandbox/implementation tenants: **$499/mo** (up to 1,000 employees), **$1,499/mo** (1,001–5,000), **$3,999/mo** (5,001–20,000), Enterprise by quote above 20,000 or for dedicated static IP / self-hosted container deployment, and **free** for one person on their own Workday developer tenant. And when employees want a visual multi-week team calendar or 3-button frontline leave screen rather than a chat thread, **[Kiweely Leaf](https://kiweely.com/leaf)** (coming next) hosts custom pass-through screens on the same user-delegated Workday security model.

## Frequently asked questions

### Why isn't an in-chat 'Type YES to confirm' prompt safe enough for Workday writes?

When confirmation happens inside the same LLM tool loop, indirect prompt injection hidden in an email, calendar invite, or pasted document can instruct the assistant to invoke the confirmation tool or misstate parameters. An out-of-band browser approval card sits outside the chat model's reach and renders the exact Workday payload directly to the human.

### What Workday API sequence is required to submit a valid time off request?

Workday's absenceManagement v5 REST service uses a multistep sequence: check balances via GET /balances, fetch eligible types via GET /workers/{ID}/eligibleAbsenceTypes, validate scheduled working days and holidays via GET /workers/{ID}/validTimeOffDates, and submit via POST /workers/{ID}/requestTimeOff with businessProcessParameters.action.id set to Submitted.

### Who appears as the initiator in Workday's business process audit trail?

When the connector uses a user-delegated Workday API client (Register API Client with Authorization Code Grant and PKCE) rather than an Integration System User, Workday records the signed-in employee as the initiator on the Request Time Off event and routes approvals through their exact supervisory chain.

### Are write actions enabled automatically when connecting Workday to Claude or ChatGPT?

In Kiweely Bridge, every tenant connection starts strictly read-only by default. Disabled write tasks are completely absent from the MCP tool list the AI assistant sees until an administrator explicitly enables specific tasks such as Request Time Off.

## Read next

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

-   [November 13 ### Preparing Workday Domain and Business Process Security Policies for a SOX Audit in Hours, Not Weeks](https://kiweely.com/blog/auditing-workday-segregation-of-duties-before-sox-review)
-   [November 4 ### The Anatomy of a Production-Safe Workday Change: Propose, Approve, Execute One by One, Read Back](https://kiweely.com/blog/anatomy-of-a-production-safe-workday-change)
-   [November 4 ### Why Workday Condition Rules Pass in Sandbox and Fail in Production](https://kiweely.com/blog/why-workday-condition-rules-fail-in-production)

### More guides on Business processes and security

[All 27 guides](https://kiweely.com/guides#business-processes)

-   [Why Workday Says ‘This Task Is Not Available’ or ‘Waiting on Initiator’ on a Business Process Step](https://kiweely.com/guides/workday-business-process-this-task-is-not-available)
-   [Why Can’t a Security Group See a Workday Dashboard or Worklet After You Added Permission?](https://kiweely.com/guides/workday-worklet-dashboard-not-visible-security)
-   [What does rescinded mean in Workday? Rescind vs cancel vs correct, and when each applies](https://kiweely.com/guides/workday-business-process-rescind-cancel-correct)
-   [Workday custom validations and condition rules: block bad data with Critical or Warning severity](https://kiweely.com/guides/workday-custom-validations-condition-rules)
-   [Workday Role-Based, User-Based, Intersection, and Aggregation Security Groups](https://kiweely.com/guides/workday-role-based-user-based-intersection-security-groups)
-   [Workday Domain Security vs. Business Process Security Policies: Which Controls What](https://kiweely.com/guides/workday-domain-vs-business-process-security-policies)
-   [Workday Business Process Advanced Routing: Exclude Initiator, Prior Approvers, and Alternates](https://kiweely.com/guides/workday-business-process-routing-mutual-exclusion)
-   [Constrained vs unconstrained security groups in Workday: which workers can a member see?](https://kiweely.com/guides/workday-constrained-vs-unconstrained-security-groups)

## Your Workday work, done by AI.

Kiweely Vine is the AI agent that builds, deploys and tests Extend apps as real people, makes changes in any Workday task screen, answers from your real tenant and builds reports.

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

## Get the next one in your inbox.

One email a month for Workday teams: what’s changing in Workday and what to retest, one tip tried in a real tenant, and what’s new at Kiweely.
