How to Opt Out of Atlassian AI Training in 20 Minutes (2026 Guide)

Matt Payne·August 16, 2026·Updated August 16, 2026·7 min read

Key Takeaway

Atlassian begins AI training on Jira and Confluence data Aug 17, 2026. An Organization Admin can disable in-app contribution in 20 minutes at admin.atlassian.com. Free, Standard, and Premium plans cannot disable metadata collection. Enterprise can.

Opt Out of Atlassian AI Training in 20 Minutes

Atlassian says data contribution starts on August 17, 2026. The policy covers Jira, Confluence, Jira Service Management, Rovo, and other Atlassian cloud apps.

"Data contribution" means Atlassian may use customer data to improve features and train AI models. You can disable some collection. Other collection stays on unless you pay for Atlassian's Enterprise plan.

Step 1: Map What Atlassian Can Collect

Atlassian separates contributed information into two groups.

Metadata includes de-identified attributes and common patterns. Examples include story points, due dates, task classifications, search patterns, and service-level values.

In-app data includes user-created content. That covers Jira titles, descriptions, comments, and custom workflow names.

It also covers Confluence page titles and page bodies.

Your Jira comments may contain customer complaints, security bugs, contract details, or product plans. Your Confluence pages may contain sales scripts, pricing plans, board notes, and employee records.

Calling that material "in-app data" doesn't lower its value.

Atlassian says it removes direct identifiers before using contributed data. The company also says it aggregates contributed data.

De-identification isn't deletion.

The reported defaults change by plan:

Atlassian plan	Metadata default	Can disable metadata?	In-app default	Can disable in-app data?
Free	On	No	On	Yes
Standard	On	No	On	Yes
Premium	On	No	Off	Yes
Enterprise	On	Yes	On	Yes

These settings apply at the organization level. They aren't separate controls for each Jira project or Confluence area.

One setting may cover engineering tickets, support cases, sales content, and internal documentation. Your marketing consultant can't review only campaign pages and call the job done.

Tool: Atlassian Administration, included with your Atlassian plan.

Expected outcome: A written map of affected apps, plans, and data categories.

Sources: Atlassian data contribution analysis and Atlassian governance review.

Step 2: Run the 20-Minute Atlassian AI Training Opt-Out

You must have the Organization Admin role.

A Product Admin may see the setting but still get an error. Atlassian's community documented that problem on July 30, 2026.

Use this script for each Atlassian organization.

Minutes 0–3: Confirm the organization and plan

1. Open admin.atlassian.com. 2. Select the correct organization. 3. Record the organization ID. 4. Record the highest active cloud plan. 5. Confirm you're an Organization Admin.

Don't assume one login means one organization. Atlassian sends separate notices for each organization.

Minutes 3–8: Open Data Contribution

Follow this path:

Atlassian Administration → Security → Data contribution

An organization-specific link may follow this format:

`https://admin.atlassian.com/o/{orgId}/data-contribution`

If the page says, "We're not sure what went wrong," check your role first. If an Organization Admin gets the same error, contact Atlassian Support.

Minutes 8–12: Disable available contribution

Turn off in-app data contribution.

Enterprise customers should also disable metadata contribution when policy requires it. Free, Standard, and Premium customers reportedly can't disable metadata contribution.

Don't promise a "full opt-out" on those three plans. It isn't available.

Minutes 12–16: Capture evidence

Take screenshots showing:

Organization name and ID
Plan level
Metadata setting
In-app data setting
Date and time
Admin account used

Save the screenshots in your security or vendor-review folder. Don't leave them in someone's Downloads directory.

Minutes 16–20: Create the review ticket

Record the control owner and next review date. Add links to Atlassian's policy, contract terms, and support case.

Set a 90-day review reminder.

Tools: Atlassian Administration and your existing ticket system. No added software is required.

Expected outcome: Contribution is disabled where available, with proof and ownership.

Official path: Atlassian's in-app data instructions.

Step 3: Audit Every AI Vendor Before Marketing Touches It

Before a marketing consultant connects an AI tool, run a vendor data-training audit.

Skipping it puts customer data at risk.

Marketing systems often hold contact lists, call transcripts, customer objections, pricing, campaign plans, and sales notes. Connecting those records before reviewing vendor terms puts convenience ahead of trust.

Atlassian's August 17 change shows why this matters. A new setting can change how a vendor uses years of stored work.

Use this checklist before connecting ChatGPT, Claude, Gemini, Rovo, Gong, HubSpot, or another AI tool.

Audit question	Low risk	Medium risk	High risk
Is customer data used for model training?	No	Optional	On by default
Can every data type be excluded?	Yes	Partial	No
Is the control organization-wide?	Clear	Mixed	Hidden
Is contributed data retained?	Under 30 days	30–90 days	Years
Can old data be removed?	Documented	Limited	Unclear
Are trained models updated after opt-out?	Documented	Delayed	No commitment
Are subprocessors named?	Yes	Partial	No
Are AI actions logged?	Full logs	Partial logs	No usable record
Can permissions be narrowed?	Per project	Per app	Full account access

Atlassian-related reports say contributed data may remain for up to seven years. They also report removal windows after an opt-out.

In-app data may take up to 30 days to remove. Metadata and model retraining may take up to 90 days.

Data already absorbed into earlier model weights may not be removed later. Check those terms against Atlassian's current contract and Trust Center.

Your audit should end with a documented decision:

Approve
Approve with restrictions
Reject
Upgrade the plan for stronger controls
Remove sensitive data before connection

Tools: Vendor terms, privacy policy, data-processing agreement, Trust Center, and admin controls.

Expected outcome: A signed decision before any consultant connects production data.

Step 4: Launch AI With Least Privilege

Opting out of vendor training doesn't make an AI agent safe.

The next risk is broad access.

An agent doesn't need every Confluence page to summarize three approved pages. It doesn't need Jira delete rights to draft ticket updates.

Create a separate service account for each agent.

Don't reuse an employee account. Don't share one token across five workflows.

Grant access only to the required Jira projects and Confluence content. Start with read-only access when possible.

Keep reading and writing separate.

A research agent can read approved records. A second workflow can submit drafted changes after human approval.

For any write action, use this control order:

1. Agent drafts the action. 2. A rule validates the fields. 3. A human approves sensitive changes. 4. A narrow credential performs the action. 5. The workflow records the result.

The rule is simple: intent can reduce permissions, never expand them. Researchers Genliang Zhu and Chu Wang call this intent-governed access control.

Don't assume SCIM manages the Data contribution setting. The supplied Atlassian sources don't confirm an API, SCIM endpoint, or supported command for that control.

Use the admin interface unless Atlassian documents another method.

Don't automate the browser with Playwright just to change a legal setting. It saves two minutes and creates a brittle control nobody trusts.

Review Atlassian audit records after connecting the agent. Keep an external log of every read, draft, write, denial, and approval.

Tools: Dedicated Atlassian account, scoped API credentials, workflow logs, and approval rules.

Expected outcome: The agent can reach only approved data and take only approved actions.

Step 5: Price the Risk Before Keeping Contribution On

The commercial decision is whether the feature value outweighs the data exposure and control costs.

Use five cost buckets:

Plan cost: Must you upgrade for metadata control?
Review cost: How much legal and security time is required?
Cleanup cost: What must be removed from Jira or Confluence?
Control cost: Who owns access, logs, and quarterly reviews?
Exit cost: How long does removal or retraining take?

Free, Standard, and Premium customers reportedly can't disable metadata contribution. Enterprise may provide that control, but Atlassian pricing must be quoted.

Some companies may accept the metadata risk. Others may decide Jira can't store certain customer details.

Write the decision down.

StoryPros builds AI agents that take action across sales, marketing, and operations. That work starts with message strategy, data boundaries, and permissions.

Start with data boundaries and permissions. Connect APIs after that.

A working agent must protect trust before it saves time. Cialdini's work on influence puts trust at the center, and AI doesn't change that.

Bad automation destroys trust faster because it runs 24/7.

Use this approval block:

> Vendor: Atlassian > Affected systems: Jira, Confluence, JSM, Rovo > Contribution setting: On / Partial / Off > Data owner: [Name] > Control owner: [Name] > Accepted risks: [List] > Required restrictions: [List] > Next review date: [Date] > Final decision: Approve / Restrict / Reject

If your AI consultant can't produce this before connecting data, find another consultant.

StoryPros offers AI consulting that ends with a working system, not a PDF. See StoryPros AI consulting for the audit and rollout process.

FAQ

Can I opt out of Atlassian AI training?

Yes. Every reported Atlassian plan can disable in-app data contribution. Only Enterprise reportedly allows a full metadata opt-out.

How do I opt out of data contribution in Atlassian?

An Organization Admin should open Atlassian Administration → Security → Data contribution. Disable in-app contribution, then disable metadata if your plan provides that control.

What Jira and Confluence data may Atlassian collect?

Metadata may include story points, dates, classifications, search patterns, and configuration values. In-app data may include Jira descriptions, comments, and Confluence page content.

How long will Atlassian retain contributed data?

Published analyses report retention of up to seven years. After you opt out, reported removal periods reach 30 days for in-app data and 90 days for metadata.

Can Free or Standard customers fully opt out?

Free and Standard customers reportedly can disable in-app contribution but not metadata contribution. Premium customers face the same metadata limit, though in-app contribution is reportedly off by default.